{
  "block": 106146770,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "7940.253069 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2026-05-18T03:02:06",
  "trx_id": "6004286f38159750052241ffe650d3d54b6e8eb8",
  "trx_in_block": 1,
  "virtual_op": 0
}

{
  "block": 105989184,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "5228.042664 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2026-05-12T15:05:39",
  "trx_id": "e6488f42e319c6f271986ff42c4694181fc29a5d",
  "trx_in_block": 3,
  "virtual_op": 0
}

{
  "block": 105514344,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "7952.768825 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2026-04-26T02:18:42",
  "trx_id": "d3ddd9ca7368c086bd143c026c1bf4154f5345be",
  "trx_in_block": 0,
  "virtual_op": 0
}

{
  "block": 102860725,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "5269.589483 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2026-01-23T15:14:24",
  "trx_id": "844d128df05aac8f7d8c99d7f6463b10a0a7c286",
  "trx_in_block": 0,
  "virtual_op": 0
}

{
  "block": 91307016,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "5433.808680 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2024-12-17T10:28:24",
  "trx_id": "539e98ad0439680d597bedbbacd807c33f48fd97",
  "trx_in_block": 2,
  "virtual_op": 0
}

{
  "block": 79861201,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "5602.942212 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2023-11-14T02:10:39",
  "trx_id": "5b53e8e322c88b07dd3eb4befaa5d7f05bcfb769",
  "trx_in_block": 0,
  "virtual_op": 0
}

{
  "block": 78351738,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "8540.220998 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2023-09-22T01:05:45",
  "trx_id": "1f0bc3b284097fe10f1998595c3f3037242a301e",
  "trx_in_block": 5,
  "virtual_op": 0
}

{
  "block": 69116584,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "8761.902436 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2022-11-03T14:29:15",
  "trx_id": "d98e02e96fdd8bf532099f485fab4a946ea3051b",
  "trx_in_block": 1,
  "virtual_op": 0
}

{
  "block": 60817566,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "8982.137572 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2022-01-17T17:46:57",
  "trx_id": "45fb8818fa30516824d46b21dc2cbc642d86a25d",
  "trx_in_block": 41,
  "virtual_op": 0
}

{
  "block": 54610717,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "9166.204325 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2021-06-14T03:19:27",
  "trx_id": "a11c4f64878708e86c12214911b39ae1e337c040",
  "trx_in_block": 3,
  "virtual_op": 0
}

{
  "block": 49358083,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "9353.626299 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2020-12-11T13:35:09",
  "trx_id": "79302acff6b8d16b0438d7a9570aab92f2756c5f",
  "trx_in_block": 13,
  "virtual_op": 0
}

{
  "block": 49209625,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "1912.543513 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2020-12-06T07:11:36",
  "trx_id": "4c0ff659b0ed809ac80a1271e223e4ce875a9fc0",
  "trx_in_block": 1,
  "virtual_op": 0
}

{
  "block": 49193172,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "9359.834153 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2020-12-05T17:13:09",
  "trx_id": "7cbe25e89eb7dd2af026c849aa07f16223bfe3eb",
  "trx_in_block": 1,
  "virtual_op": 0
}

{
  "block": 48263804,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "1920.017158 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2020-11-02T20:44:09",
  "trx_id": "8c96b12f415731105f1122a9e34c4bfa340a7ad4",
  "trx_in_block": 4,
  "virtual_op": 0
}

{
  "block": 43219910,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "9562.639512 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2020-05-09T08:11:42",
  "trx_id": "39ba5fc0512b95f125b2b152f612c5903c18cc82",
  "trx_in_block": 0,
  "virtual_op": 0
}

{
  "block": 43196444,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "1953.311140 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2020-05-08T12:10:00",
  "trx_id": "40c9391f2bffad8a966afb68f1123d8eefa5e451",
  "trx_in_block": 0,
  "virtual_op": 0
}

{
  "block": 35408208,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "9718.764135 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2019-08-09T18:17:27",
  "trx_id": "1a51043fce22e66d3d8416c91de265d453803433",
  "trx_in_block": 34,
  "virtual_op": 0
}

{
  "block": 33178282,
  "op": [
    "comment",
    {
      "author": "steemitboard",
      "body": "Congratulations @llsourcell! You received a personal award!\n\n<table><tr><td>https://steemitimages.com/70x70/http://steemitboard.com/@llsourcell/birthday1.png</td><td>Happy Birthday! - You are on the Steem blockchain for 1 year!</td></tr></table>\n\n<sub>_You can view [your badges on your Steem Board](https://steemitboard.com/@llsourcell) and compare to others on the [Steem Ranking](http://steemitboard.com/ranking/index.php?name=llsourcell)_</sub>\n\n\n###### [Vote for @Steemitboard as a witness](https://v2.steemconnect.com/sign/account-witness-vote?witness=steemitboard&approve=1) to get one more award and increased upvotes!",
      "json_metadata": "{\"image\":[\"https://steemitboard.com/img/notify.png\"]}",
      "parent_author": "llsourcell",
      "parent_permlink": "725d78b0-6062-11e8-b143-ffce16c65548",
      "permlink": "steemitboard-notify-llsourcell-20190524t042438000z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2019-05-24T04:24:39",
  "trx_id": "b610226ed09b13af8f4bfa131f0e774f24f66ee3",
  "trx_in_block": 2,
  "virtual_op": 0
}

{
  "block": 25361282,
  "op": [
    "delegate_vesting_shares",
    {
      "delegatee": "llsourcell",
      "delegator": "steem",
      "vesting_shares": "9917.256687 VESTS"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-08-24T23:24:06",
  "trx_id": "11bc5745836efc6a008e54d53d2f39eeb0677c0a",
  "trx_in_block": 16,
  "virtual_op": 0
}

{
  "block": 23065230,
  "op": [
    "comment",
    {
      "author": "guiltyparties",
      "body": "!cheetah ban\n\nID thief - no appeal.",
      "json_metadata": "{\"tags\":[\"dlive\"],\"app\":\"steemit/0.1\"}",
      "parent_author": "llsourcell",
      "parent_permlink": "725d78b0-6062-11e8-b143-ffce16c65548",
      "permlink": "re-llsourcell-725d78b0-6062-11e8-b143-ffce16c65548-20180605t201208102z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-06-05T20:11:42",
  "trx_id": "fa1bf671fa39d1ddc842dc1f978b94535b51aff3",
  "trx_in_block": 19,
  "virtual_op": 0
}

{
  "block": 22925192,
  "op": [
    "comment",
    {
      "author": "steemcleaners",
      "body": "Hello, \n\nWe have contacted you on your Twitter to verify the authorship of your Steemit blog but we have received no response yet. We would be grateful if you could, please respond to us via Twitter. \n\nhttps://twitter.com/steemcleaners/status/1001570790054252544\n\nPlease note I am a volunteer that works to ensure that plagiarised content does not get rewarded. I have no way to remove any content from steemit.com.\n\nThank you",
      "json_metadata": "{\"tags\":[\"dlive\"],\"links\":[\"https://twitter.com/steemcleaners/status/1001570790054252544\"],\"app\":\"steemit/0.1\"}",
      "parent_author": "llsourcell",
      "parent_permlink": "725d78b0-6062-11e8-b143-ffce16c65548",
      "permlink": "re-llsourcell-725d78b0-6062-11e8-b143-ffce16c65548-20180531t232640179z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-31T23:26:42",
  "trx_id": "8dc5d6e72b23687a6c5ecef325f418d42c67346b",
  "trx_in_block": 15,
  "virtual_op": 0
}

{
  "block": 22751329,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "introducemyself-welcome-steemit-folks",
      "voter": "cheneats",
      "weight": 100
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T22:32:21",
  "trx_id": "f5394ac32b759e74d85036eadbff0195848127ae",
  "trx_in_block": 66,
  "virtual_op": 0
}

{
  "block": 22751322,
  "op": [
    "comment",
    {
      "author": "cheneats",
      "body": "Welcome to Steemit, llsourcell! Wish you a very fun journey here on this platform :) Have fun!!\n\nBy the way, there are several groups you as a newcomer can join. They will stay with you for your journey, helping and mentoring along the way.\n\n@greetersguild invite link https://discord.gg/AkzNSKx\n@newbieresteemday invite link https://discord.gg/2ZcAxsU",
      "json_metadata": "",
      "parent_author": "llsourcell",
      "parent_permlink": "introducemyself-welcome-steemit-folks",
      "permlink": "re-introducemyself-welcome-steemit-folks-20180525t223200",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T22:32:00",
  "trx_id": "ee2bb6d80da6514c0594d302e9f34c10bd1c0ce6",
  "trx_in_block": 44,
  "virtual_op": 0
}

{
  "block": 22750170,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "725d78b0-6062-11e8-b143-ffce16c65548",
      "voter": "ubg",
      "weight": 100
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T21:34:21",
  "trx_id": "1a30520ad94a21cfc1297fc67e19460db4e9bd7b",
  "trx_in_block": 16,
  "virtual_op": 0
}

{
  "block": 22750148,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "[![Thumbnail](https://dlive.biz:8080/QmVyiAjNmDTkMvdpgCemfF5cN8RnbbAkx9aFUi7ATrj4gr)](https://dlive.io/video/llsourcell/725d78b0-6062-11e8-b143-ffce16c65548)\n\nSelf driving cars are the eventual fate of transportation and will make up a significant piece of society as more drive related employments are robotized. In this video, i'll clarify how the whole self driving auto pipeline works, including PC vision, way arranging, control, sensor combination, and limitation. We'll utilize the Udacity test system to prepare our own self driving auto with the Keras profound learning library as an instrument toward the end. This innovation is shockingly easy to comprehend, it just requires look into two or three subfields, all of which i'll cover. \n\n# Code for this video: \n\nhttps://github.com/llSourcell/\n\nMy video is at [DLive](https://dlive.io/video/llsourcell/725d78b0-6062-11e8-b143-ffce16c65548)",
      "json_metadata": "{\"tags\":[\"dlive\",\"dlive-video\",\"game\",\"technology \",\"programming \"],\"app\":\"dlive/0.1\",\"format\":\"markdown\",\"language\":\"en\",\"thumbnail\":\"https://dlive.biz:8080/QmVyiAjNmDTkMvdpgCemfF5cN8RnbbAkx9aFUi7ATrj4gr\",\"ipfsHash\":\"QmXbq3bbbQmx8gWw7Sdq42z5zr6jSkK2vckbZNzc8xUTMK\"}",
      "parent_author": "",
      "parent_permlink": "dlive",
      "permlink": "725d78b0-6062-11e8-b143-ffce16c65548",
      "title": "SelfDriving Cars Explained"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T21:33:15",
  "trx_id": "a265577b8d57f981e101831ccc3d1dee07bc3e3a",
  "trx_in_block": 65,
  "virtual_op": 0
}

{
  "block": 22749717,
  "op": [
    "comment",
    {
      "author": "cheetah",
      "body": "Hi! I am a robot. I just upvoted you! I found similar content that readers might be interested in:\nhttps://github.com/salu133445/musegan",
      "json_metadata": "",
      "parent_author": "llsourcell",
      "parent_permlink": "t4dy0sxk",
      "permlink": "cheetah-re-llsourcellt4dy0sxk",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T21:11:42",
  "trx_id": "6c013facc300cb20f6f6a3589f47b6109f728fe0",
  "trx_in_block": 41,
  "virtual_op": 0
}

{
  "block": 22749708,
  "op": [
    "comment_options",
    {
      "allow_curation_rewards": true,
      "allow_votes": true,
      "author": "llsourcell",
      "extensions": [
        [
          0,
          {
            "beneficiaries": [
              {
                "account": "dtube",
                "weight": 2500
              }
            ]
          }
        ]
      ],
      "max_accepted_payout": "1000000.000 SBD",
      "percent_steem_dollars": 10000,
      "permlink": "t4dy0sxk"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T21:11:15",
  "trx_id": "412ea8e59628d9b5494b75acff898f0e6dd2b42c",
  "trx_in_block": 53,
  "virtual_op": 0
}

{
  "block": 22749708,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "<center><a href='https://d.tube/#!/v/llsourcell/t4dy0sxk'><img src='https://ipfs.io/ipfs/QmbfTSpgYRhBG4Jc6kzmeJQ2jjGbRf7Rb9JCRrTApHi1Kn'></a></center><hr>\n\n# MuseGAN\n\nMuseGAN is a project on music generation. In essence, we aim to generate polyphonic music of multiple tracks (instruments) with harmonic and rhythmic structure, multi-track interdependency and temporal structure. To our knowledge, our work represents the first approach that deal with these issues altogether.\n\nThe models are trained with Lakh Pianoroll Dataset (LPD), a new multi-track piano-roll dataset, in an unsupervised approach. The proposed models are able to generate music either from scratch, or by accompanying a track given by user. Specifically, we use the model to generate pop song phrases consisting of bass, drums, guitar, piano and strings tracks.\n\nSample results are available [here](https://salu133445.github.io/musegan/results).\n\n# BinaryMuseGAN\n\nBinaryMuseGAN is a follow-up project of the MuseGAN project.\n\nIn this project, we first investigate how the real-valued piano-rolls generated by the generator may lead to difficulties in training the discriminator for CNN-based models. To overcome the binarization issue, we propose to append to the generator an additional refiner network, which try to refine the real-valued predictions generated by the pretrained generator to binary-valued ones. The proposed model is able to directly generate binary-valued piano-rolls at test time.\n\nWe trained the network with Lakh Pianoroll Dataset (LPD). We use the model to generate four-bar musical phrases consisting of eight tracks: Drums, Piano, Guitar, Bass, Ensemble, Reed, Synth Lead and Synth Pad. Audio samples are available [here](https://salu133445.github.io/bmusegan/samples).\n\n## Run the code\n\nPrepare Training Data\n\nPrepare your own data or download our training data\n\n> The array will be reshaped to (-1, num_bar, num_timestep, num_pitch, num_track). These variables are defined in config.py.\n\n> lastfm_alternative_5b_phrase.npy (2.1 GB) contains 12,444 four-bar phrases from 2,074 songs with alternative tags. The shape is (2074, 6, 4, 96, 84, 5). The five tracks are Drums, Piano, Guitar, Bass and Strings.\nlastfm_alternative_8b_phrase.npy (3.6 GB) contains 13,746 four-bar phrases from 2,291 songs with alternative tags. The shape is (2291, 6, 4, 96, 84, 8). The eight tracks are Drums, Piano, Guitar, Bass, Ensemble, Reed, Synth Lead and Synth Pad.\n\nDownload the data with this script.\n(optional) Save the training data to shared memory with this script.\n\nSpecify training data path and location in config.py. (see below)\n\n## Configuration\n\n    Modify config.py for configuration.\n\n## Quick setup\n\nChange the values in the dictionary SETUP for a quick setup. Documentation is provided right after each key.\n\n## More configuration options\n\n> Four dictionaries EXP_CONFIG, DATA_CONFIG, MODEL_CONFIG and TRAIN_CONFIG define experiment-, data-, model- and training-related configuration variables, respectively.\n\nThe automatically-determined experiment name is based only on the values defined in the dictionary SETUP, so remember to provide the experiment name manually (so that you won't overwrite a trained model).\n\n## Run\n\n      python main.py\n\n# Github Repository :\n\n* https://github.com/llSourcell/AI_For_Music_Composition\n\n\n<hr><a href='https://d.tube/#!/v/llsourcell/t4dy0sxk'> ▶️ DTube</a><br /><a href='https://ipfs.io/ipfs/QmNm1snxssQeLwTD3wqev7dytcEkoXEA1eZW9S6wmiz3oP'> ▶️ IPFS</a>",
      "json_metadata": "{\"video\":{\"info\":{\"title\":\"AI for Music Composition\",\"snaphash\":\"QmSqR2Wzk15D2XZpvafLYT77XFsrC8t9Md3RsQuRJQgn1h\",\"author\":\"llsourcell\",\"permlink\":\"t4dy0sxk\",\"duration\":1309.048163,\"filesize\":109163977,\"spritehash\":\"QmS7apEHhmEGEDAC8qyRhVpkn31araBoG57eUyur2AC7f8\"},\"content\":{\"videohash\":\"QmNm1snxssQeLwTD3wqev7dytcEkoXEA1eZW9S6wmiz3oP\",\"description\":\"# MuseGAN\\n\\nMuseGAN is a project on music generation. In essence, we aim to generate polyphonic music of multiple tracks (instruments) with harmonic and rhythmic structure, multi-track interdependency and temporal structure. To our knowledge, our work represents the first approach that deal with these issues altogether.\\n\\nThe models are trained with Lakh Pianoroll Dataset (LPD), a new multi-track piano-roll dataset, in an unsupervised approach. The proposed models are able to generate music either from scratch, or by accompanying a track given by user. Specifically, we use the model to generate pop song phrases consisting of bass, drums, guitar, piano and strings tracks.\\n\\nSample results are available [here](https://salu133445.github.io/musegan/results).\\n\\n# BinaryMuseGAN\\n\\nBinaryMuseGAN is a follow-up project of the MuseGAN project.\\n\\nIn this project, we first investigate how the real-valued piano-rolls generated by the generator may lead to difficulties in training the discriminator for CNN-based models. To overcome the binarization issue, we propose to append to the generator an additional refiner network, which try to refine the real-valued predictions generated by the pretrained generator to binary-valued ones. The proposed model is able to directly generate binary-valued piano-rolls at test time.\\n\\nWe trained the network with Lakh Pianoroll Dataset (LPD). We use the model to generate four-bar musical phrases consisting of eight tracks: Drums, Piano, Guitar, Bass, Ensemble, Reed, Synth Lead and Synth Pad. Audio samples are available [here](https://salu133445.github.io/bmusegan/samples).\\n\\n## Run the code\\n\\nPrepare Training Data\\n\\nPrepare your own data or download our training data\\n\\n> The array will be reshaped to (-1, num_bar, num_timestep, num_pitch, num_track). These variables are defined in config.py.\\n\\n> lastfm_alternative_5b_phrase.npy (2.1 GB) contains 12,444 four-bar phrases from 2,074 songs with alternative tags. The shape is (2074, 6, 4, 96, 84, 5). The five tracks are Drums, Piano, Guitar, Bass and Strings.\\nlastfm_alternative_8b_phrase.npy (3.6 GB) contains 13,746 four-bar phrases from 2,291 songs with alternative tags. The shape is (2291, 6, 4, 96, 84, 8). The eight tracks are Drums, Piano, Guitar, Bass, Ensemble, Reed, Synth Lead and Synth Pad.\\n\\nDownload the data with this script.\\n(optional) Save the training data to shared memory with this script.\\n\\nSpecify training data path and location in config.py. (see below)\\n\\n## Configuration\\n\\n    Modify config.py for configuration.\\n\\n## Quick setup\\n\\nChange the values in the dictionary SETUP for a quick setup. Documentation is provided right after each key.\\n\\n## More configuration options\\n\\n> Four dictionaries EXP_CONFIG, DATA_CONFIG, MODEL_CONFIG and TRAIN_CONFIG define experiment-, data-, model- and training-related configuration variables, respectively.\\n\\nThe automatically-determined experiment name is based only on the values defined in the dictionary SETUP, so remember to provide the experiment name manually (so that you won't overwrite a trained model).\\n\\n## Run\\n\\n      python main.py\\n\\n# Github Repository :\\n\\n* https://github.com/llSourcell/AI_For_Music_Composition\\n\",\"tags\":[\"dtube\",\"technology\",\"music\"],\"video480hash\":\"QmRujYZkJ585XBK6vCP3v5pzvnevBbaXRc1nSsqD1Yqykr\"}},\"tags\":[\"dtube\",\"technology\",\"music\",\"dtube\"],\"app\":\"dtube/0.7\"}",
      "parent_author": "",
      "parent_permlink": "dtube",
      "permlink": "t4dy0sxk",
      "title": "AI for Music Composition"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T21:11:15",
  "trx_id": "412ea8e59628d9b5494b75acff898f0e6dd2b42c",
  "trx_in_block": 53,
  "virtual_op": 0
}

{
  "block": 22737470,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "bwvheqq3",
      "voter": "sultanmr",
      "weight": 10000
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T10:59:06",
  "trx_id": "818c5f0cab67c2c7a8b0dcbacfb67e4966d89f1e",
  "trx_in_block": 12,
  "virtual_op": 0
}

{
  "block": 22737469,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "introducemyself-welcome-steemit-folks",
      "voter": "sultanmr",
      "weight": 10000
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T10:59:03",
  "trx_id": "787eb8b7d97c3024f8ecf867f14858aedc630b72",
  "trx_in_block": 45,
  "virtual_op": 0
}

{
  "block": 22735992,
  "op": [
    "account_update",
    {
      "account": "llsourcell",
      "json_metadata": "{\"profile\":{\"profile_image\":\"http://i66.tinypic.com/s4vq0y.png\",\"name\":\"Siraj Raval\",\"about\":\"Director at School of AI, inspire and educate developers to build AI.\",\"location\":\"San Francisco, CA\",\"website\":\"https://www.youtube.com/c/sirajraval\"}}",
      "memo_key": "STM7tn6mUDrHTLfT86EPoNGAM36NVSRx7hy7fKBDApRv1pwrMaXu8",
      "posting": {
        "account_auths": [
          [
            "dlive.app",
            1
          ],
          [
            "dtube.app",
            1
          ]
        ],
        "key_auths": [
          [
            "STM5coHQ3vpih6hqZTWFnaUbUG7MZqhBjWoPTHyErEN7xdmJWDX8p",
            1
          ]
        ],
        "weight_threshold": 1
      }
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T09:45:12",
  "trx_id": "9282a976d8b8400dec731a5d59c9ec782f9ea25c",
  "trx_in_block": 39,
  "virtual_op": 0
}

{
  "block": 22734976,
  "op": [
    "comment_options",
    {
      "allow_curation_rewards": true,
      "allow_votes": true,
      "author": "llsourcell",
      "extensions": [
        [
          0,
          {
            "beneficiaries": [
              {
                "account": "dtube",
                "weight": 2500
              }
            ]
          }
        ]
      ],
      "max_accepted_payout": "1000000.000 SBD",
      "percent_steem_dollars": 10000,
      "permlink": "bwvheqq3"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T08:54:24",
  "trx_id": "b75ef9917ca06575ad9d5e3a7dda2f9707009a1e",
  "trx_in_block": 29,
  "virtual_op": 0
}

{
  "block": 22734976,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "<center><a href='https://d.tube/#!/v/llsourcell/bwvheqq3'><img src='https://ipfs.io/ipfs/Qmb6VAiWqemPNpBGJXoQKzWphwT76hPckcAz3pckCP2a7g'></a></center><hr>\n\nI'm introducing a cryptocurrency for our community called SirajCoin. It will act as the fuel for our global community of developers to engage with me directly by spending it on my attention via hourly meetings and video collaborations. \n\n# Sirajcoin\n\nHello world, it's Sirajcoin!\n\nSirajcoin is an experiment with two goals:\n\n* add rocket fuel to the growth of our community\n* fund AI research in a decentralized way!\n\n**Note**: this is still very experimental code, and it may be insecure. Please do not pay real money for Sirajcoin.\n\nInstalling the Sirajcoin wallet\nTo send and receive Sirajcoin, you'll need Node.js version 8 or later.\n\nThen in your terminal, run:\n\n     npm i -g sirajcoin\n\nIf you're having trouble installing, try installing it locally:\n\n    mkdir -p ~/.sirajcoin && \\\n   cd ~/.sirajcoin && \\\n    echo {} > package.json && \\\n     npm i sirajcoin && \\\n     export \n    PATH=$PATH:$PWD/\n    node_modules/.bin\n \n# Getting your first Sirajcoin\n\nFirst, you'll need to generate a Sirajcoin address.\n\nAfter you've installed the wallet, you can see your address by running:\n\n     $ sirajcoin balance\n\nTo get your first Sirajcoin, simply subscribe to the [YouTube channel](https://www.youtube.com/channel/UCWN3xxRkmTPmbKwht9FuE5A?app=desktop), then post a comment containing your Sirajcoin address on this video, and the Sirajcoin YouTube oracle will automagically grant you 10 Sirajcoin if you've subscribed.\n\nSending Sirajcoin\nTo send Sirajcoin to someone, run:\n\n    $ sirajcoin send <recipient_address> \n   <amount>\n\n\n#How it works\n\nIn short, Sirajcoin is a proof-of-stake cryptocurrency built on top of Tendermint consensus using a library called Lotion, secured by a hand-picked set of community validators.\n\nYou can learn more about the technical details and economic design of Sirajcoin\n in the Sirajcoin whitepaper.\n\n# Credits\n\nSirajcoin is developed by:\n\n* Siraj Raval\n* Matt Bell\n* Chad Lohrli\n* Judd Keppel\n* Anders Thuesen\n\nand you are encouraged to contribute ideas or pull requests!\n\n# My Github Repository: \n* https://github.com/llSourcell/sirajcoin\n\n\n<hr><a href='https://d.tube/#!/v/llsourcell/bwvheqq3'> ▶️ DTube</a><br /><a href='https://ipfs.io/ipfs/QmaeorALLaoJ17qmkEhSTqirdN5zkQ4VGwikpi8QErhLDL'> ▶️ IPFS</a>",
      "json_metadata": "{\"video\":{\"info\":{\"title\":\"SirajCoin The Cryptpcurrency of Programming Wizards\",\"snaphash\":\"Qmb9dVLjhrycVZKHG1uv25FTwqqboAYzJsrofZaSKAVMaF\",\"author\":\"llsourcell\",\"permlink\":\"bwvheqq3\",\"duration\":673.680544,\"filesize\":50061918,\"spritehash\":\"QmSPuhcooxoU82dz4rFSnVgarF5Qm7e9oLJ8CVUYgMfvvh\"},\"content\":{\"videohash\":\"QmaeorALLaoJ17qmkEhSTqirdN5zkQ4VGwikpi8QErhLDL\",\"description\":\"I'm introducing a cryptocurrency for our community called SirajCoin. It will act as the fuel for our global community of developers to engage with me directly by spending it on my attention via hourly meetings and video collaborations. \\n\\n# Sirajcoin\\n\\nHello world, it's Sirajcoin!\\n\\nSirajcoin is an experiment with two goals:\\n\\n* add rocket fuel to the growth of our community\\n* fund AI research in a decentralized way!\\n\\n**Note**: this is still very experimental code, and it may be insecure. Please do not pay real money for Sirajcoin.\\n\\nInstalling the Sirajcoin wallet\\nTo send and receive Sirajcoin, you'll need Node.js version 8 or later.\\n\\nThen in your terminal, run:\\n\\n     npm i -g sirajcoin\\n\\nIf you're having trouble installing, try installing it locally:\\n\\n    mkdir -p ~/.sirajcoin && \\\\\\n   cd ~/.sirajcoin && \\\\\\n    echo {} > package.json && \\\\\\n     npm i sirajcoin && \\\\\\n     export \\n    PATH=$PATH:$PWD/\\n    node_modules/.bin\\n \\n# Getting your first Sirajcoin\\n\\nFirst, you'll need to generate a Sirajcoin address.\\n\\nAfter you've installed the wallet, you can see your address by running:\\n\\n     $ sirajcoin balance\\n\\nTo get your first Sirajcoin, simply subscribe to the [YouTube channel](https://www.youtube.com/channel/UCWN3xxRkmTPmbKwht9FuE5A?app=desktop), then post a comment containing your Sirajcoin address on this video, and the Sirajcoin YouTube oracle will automagically grant you 10 Sirajcoin if you've subscribed.\\n\\nSending Sirajcoin\\nTo send Sirajcoin to someone, run:\\n\\n    $ sirajcoin send <recipient_address> \\n   <amount>\\n\\n\\n#How it works\\n\\nIn short, Sirajcoin is a proof-of-stake cryptocurrency built on top of Tendermint consensus using a library called Lotion, secured by a hand-picked set of community validators.\\n\\nYou can learn more about the technical details and economic design of Sirajcoin\\n in the Sirajcoin whitepaper.\\n\\n# Credits\\n\\nSirajcoin is developed by:\\n\\n* Siraj Raval\\n* Matt Bell\\n* Chad Lohrli\\n* Judd Keppel\\n* Anders Thuesen\\n\\nand you are encouraged to contribute ideas or pull requests!\\n\\n# My Github Repository: \\n* https://github.com/llSourcell/sirajcoin\\n\",\"tags\":[\"sirajcoin\",\"dtubedaily\",\"onelovedtube\"],\"video480hash\":\"QmYQodRs4c1SZJ9KGS1pCjwh8qJgni37jZG5uVGDvyuN1a\"}},\"tags\":[\"sirajcoin\",\"dtubedaily\",\"onelovedtube\",\"dtube\"],\"app\":\"dtube/0.7\"}",
      "parent_author": "",
      "parent_permlink": "sirajcoin",
      "permlink": "bwvheqq3",
      "title": "SirajCoin The Cryptpcurrency of Programming Wizards"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T08:54:24",
  "trx_id": "b75ef9917ca06575ad9d5e3a7dda2f9707009a1e",
  "trx_in_block": 29,
  "virtual_op": 0
}

{
  "block": 22734631,
  "op": [
    "account_update",
    {
      "account": "llsourcell",
      "json_metadata": "{\"profile\":{\"profile_image\":\"http://i66.tinypic.com/s4vq0y.png\",\"name\":\"Siraj Raval\",\"about\":\"Director at School of AI, inspire and educate developers to build AI.\",\"location\":\"San Francisco, CA\",\"website\":\"https://www.youtube.com/c/sirajraval\"}}",
      "memo_key": "STM7tn6mUDrHTLfT86EPoNGAM36NVSRx7hy7fKBDApRv1pwrMaXu8",
      "posting": {
        "account_auths": [
          [
            "dtube.app",
            1
          ]
        ],
        "key_auths": [
          [
            "STM5coHQ3vpih6hqZTWFnaUbUG7MZqhBjWoPTHyErEN7xdmJWDX8p",
            1
          ]
        ],
        "weight_threshold": 1
      }
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T08:37:09",
  "trx_id": "1ef0eb421dddbe86b786d559e9889e4addbc9a12",
  "trx_in_block": 6,
  "virtual_op": 0
}

{
  "block": 22729079,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "self-driving-cars-explained",
      "voter": "sultanmr",
      "weight": 10000
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T03:59:30",
  "trx_id": "902e35fdfc18950c104cea557fdebabdf77a5599",
  "trx_in_block": 54,
  "virtual_op": 0
}

{
  "block": 22729076,
  "op": [
    "comment",
    {
      "author": "sultanmr",
      "body": "siraj bhaya, im big fan of yours. you always create a very impressive youtube videos, yaar yahan steem pay just typing blog of your level, you will get nothing as it worth, because yahan bhot say fazool stressfull bots chulay howay hain. agr aap simply apnay youtube kay video d.tube pay upload kur dain, jo kay steem say link hai, then you will get good rewards without doing anything extra here. its good to see you here, but whatever you are doing, is really great, so this is just a suggestion to keep on doing that and u will get more reward ultimatelly. just google d.tube plz\nthanks",
      "json_metadata": "{\"tags\":[\"utopian-io\"],\"app\":\"steemit/0.1\"}",
      "parent_author": "llsourcell",
      "parent_permlink": "self-driving-cars-explained",
      "permlink": "re-llsourcell-self-driving-cars-explained-20180525t035919766z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-25T03:59:21",
  "trx_id": "54af90cb2b756909f2598da8534e8671bd9d8cba",
  "trx_in_block": 3,
  "virtual_op": 0
}

{
  "block": 22708440,
  "op": [
    "comment",
    {
      "author": "mcfarhat",
      "body": "Plagiarizing content is a serious offense. Your content is available online on this link https://github.com/upul/Behavioral-Cloning\nYou have been banned from receiving Utopian reviews for 60 days. \nSimilar contributions in the future would lead to permanent ban.\n\n---- \nNeed help? Write a ticket on https://support.utopian.io/. \nChat with us on [Discord](https://discord.gg/uTyJkNm). \n[[utopian-moderator]](https://join.utopian.io/)",
      "json_metadata": "{\"tags\":[\"utopian-io\"],\"links\":[\"https://github.com/upul/Behavioral-Cloning\",\"https://support.utopian.io/\",\"https://discord.gg/uTyJkNm\",\"https://join.utopian.io/\"],\"app\":\"steemit/0.1\"}",
      "parent_author": "llsourcell",
      "parent_permlink": "self-driving-cars-explained",
      "permlink": "re-llsourcell-self-driving-cars-explained-20180524t103810896z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T10:38:18",
  "trx_id": "6b529c2dbf0f8d50d2ba815dc5bd76ba54473404",
  "trx_in_block": 3,
  "virtual_op": 0
}

{
  "block": 22708351,
  "op": [
    "comment",
    {
      "author": "funb",
      "body": "nope @siraj it will also considered as plagiarism",
      "json_metadata": "{\"tags\":[\"utopian-io\"],\"users\":[\"siraj\"],\"app\":\"steemit/0.1\"}",
      "parent_author": "llsourcell",
      "parent_permlink": "re-funb-re-llsourcell-self-driving-cars-explained-20180524t103351538z",
      "permlink": "re-llsourcell-re-funb-re-llsourcell-self-driving-cars-explained-20180524t103325062z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T10:33:33",
  "trx_id": "665c94d8f89b3f7a88ab9573e76a0f2724c66edd",
  "trx_in_block": 20,
  "virtual_op": 0
}

{
  "block": 22708313,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "Thank you @funb.\n\nI have written some post on medium.com.  Can I repost here which I have written?",
      "json_metadata": "{\"tags\":[\"utopian-io\"],\"users\":[\"funb\"],\"app\":\"steemit/0.1\"}",
      "parent_author": "funb",
      "parent_permlink": "re-llsourcell-self-driving-cars-explained-20180524t101148850z",
      "permlink": "re-funb-re-llsourcell-self-driving-cars-explained-20180524t103351538z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T10:31:33",
  "trx_id": "8ba06915f9874e5edb85cf532023ec06dbe547ab",
  "trx_in_block": 43,
  "virtual_op": 0
}

{
  "block": 22708000,
  "op": [
    "comment",
    {
      "author": "funb",
      "body": "@siraj i think that you are a new at steemit and don't know the utopian rules. Utopian doesn't allow any person to copy any other person material and then paste in your post, and in utopian this is known as plagiarism. And due to this reason you would be ban at utopian.And would be not able to write more post at utopian.\n![](https://cdn.steemitimages.com/DQmNmQRxBPv9byUhvVvoPVqVEtxeVWooxEPDoPYrq2N6hBT/image.png)\n![](https://cdn.steemitimages.com/DQmYKpj7mdu712GfyuXeWyfMm6tPMDaAsgZKU64MZSsdWaW/image.png)",
      "json_metadata": "{\"tags\":[\"utopian-io\"],\"users\":[\"siraj\"],\"image\":[\"https://cdn.steemitimages.com/DQmNmQRxBPv9byUhvVvoPVqVEtxeVWooxEPDoPYrq2N6hBT/image.png\",\"https://cdn.steemitimages.com/DQmYKpj7mdu712GfyuXeWyfMm6tPMDaAsgZKU64MZSsdWaW/image.png\"],\"app\":\"steemit/0.1\"}",
      "parent_author": "llsourcell",
      "parent_permlink": "self-driving-cars-explained",
      "permlink": "re-llsourcell-self-driving-cars-explained-20180524t101148850z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T10:15:00",
  "trx_id": "42ff691a441300f442efd8aa34fa52647ee11785",
  "trx_in_block": 87,
  "virtual_op": 0
}

{
  "block": 22707942,
  "op": [
    "comment",
    {
      "author": "funb",
      "body": "@siraj i think that you are a new at steemit and don't know the utopian rules. Utopian doesn't allow any person to copy any other person material and then paste in your post, and in utopian this is known as plagiarism. And due to this reason you would be ban at utopian.And would be not able to write more post at utopian.\n![](https://cdn.steemitimages.com/DQmNmQRxBPv9byUhvVvoPVqVEtxeVWooxEPDoPYrq2N6hBT/image.png)\n![](https://cdn.steemitimages.com/DQmYKpj7mdu712GfyuXeWyfMm6tPMDaAsgZKU64MZSsdWaW/image.png)",
      "json_metadata": "{\"tags\":[\"utopian-io\"],\"users\":[\"siraj\"],\"image\":[\"https://cdn.steemitimages.com/DQmNmQRxBPv9byUhvVvoPVqVEtxeVWooxEPDoPYrq2N6hBT/image.png\",\"https://cdn.steemitimages.com/DQmYKpj7mdu712GfyuXeWyfMm6tPMDaAsgZKU64MZSsdWaW/image.png\"],\"app\":\"steemit/0.1\"}",
      "parent_author": "llsourcell",
      "parent_permlink": "self-driving-cars-explained",
      "permlink": "re-llsourcell-self-driving-cars-explained-20180524t101148850z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T10:11:57",
  "trx_id": "4a21a1b49747e60fc6257c7afe7be52ca99e2815",
  "trx_in_block": 12,
  "virtual_op": 0
}

{
  "block": 22705898,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "@@ -2553,16 +2553,35 @@\n ..%3C/h3%3E%0A\n+%3Cp%3EVideo %3C/p%3E%0A\n %3Cp%3Ehttps\n",
      "json_metadata": "{\"tags\":[\"utopian-io\",\"video-tutorials\",\"development\",\"technology\"],\"image\":[\"http://i65.tinypic.com/2cdfkmx.jpg\",\"https://img.youtube.com/vi/yt015gM-ync/0.jpg\"],\"links\":[\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \",\"https://github.com/udacity/self-driving-car-sim\",\"https://keras.io\",\"www.numpy.org\",\"https://www.scipy.org\",\"https://www.tensorflow.org\",\"pandas.pydata.org\",\"https://opencv.org\",\"https://matplotlib.org\",\"jupyter.org\",\"https://github.com/llSourcell/self_driving_cars_explained?files=1\",\"https://www.youtube.com/watch?v=yt015gM-ync&feature=youtu.be\",\"https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI\",\"https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c\",\"https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8\",\"https://searchenterpriseai.techtarget.com/definition/driverless-car\",\"https://github.com/llSourcell/self_driving_cars_explained\"],\"app\":\"steemit/0.1\",\"format\":\"html\"}",
      "parent_author": "",
      "parent_permlink": "utopian-io",
      "permlink": "self-driving-cars-explained",
      "title": "Self-Driving Cars Explained"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T08:22:15",
  "trx_id": "3e321b571513885a458cf2c099b946e8c2e75ae0",
  "trx_in_block": 79,
  "virtual_op": 0
}

{
  "block": 22705894,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "self-driving-cars-explained",
      "voter": "smartmediagroup",
      "weight": 225
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T08:22:03",
  "trx_id": "d24326b3c4d8efd54b31df1dba2a78d7f6b63130",
  "trx_in_block": 0,
  "virtual_op": 0
}

{
  "block": 22705875,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "@@ -927,50 +927,8 @@\n /p%3E%0A\n-%3Cp%3Ehttp://i66.tinypic.com/2a8qgxy.jpg%3C/p%3E%0A\n %3Ch1%3E\n@@ -1027,16 +1027,310 @@\n %3E%0A%3C/ul%3E%0A\n+%3Ch1%3EWhat Will I Learn? %3C/h1%3E%0A%3Cul%3E%0A  %3Cli%3EYou learn how the entire self driving car pipeline works%3C/li%3E%0A  %3Cli%3E You Learn computer visioin, path planning%3C/li%3E%0A  %3Cli%3EYou Learn control, sensor fusion and localization.%3C/li%3E%0A%3C/ul%3E%0A%3Ch1%3EDifficulty %3C/h1%3E%0A%3Cul%3E%0A  %3Cli%3EBasic %3C/li%3E%0A%3C/ul%3E%0A\n %3Ch1%3ERequ\n@@ -3513,24 +3513,220 @@\n %3C/li%3E%0A%3C/ul%3E%0A\n+%3Ch2%3ESummary %3C/h2%3E%0A%3Cp%3EIn this tutorial, I explain how the entire self driving car pipeline works, including computer vision, path planning, control, sensor fusion, and localization. %3C/p%3E%0A\n %3Ch2%3EMy Repos\n",
      "json_metadata": "{\"tags\":[\"utopian-io\",\"video-tutorials\",\"development\",\"technology\"],\"image\":[\"http://i65.tinypic.com/2cdfkmx.jpg\",\"https://img.youtube.com/vi/yt015gM-ync/0.jpg\"],\"links\":[\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \",\"https://github.com/udacity/self-driving-car-sim\",\"https://keras.io\",\"www.numpy.org\",\"https://www.scipy.org\",\"https://www.tensorflow.org\",\"pandas.pydata.org\",\"https://opencv.org\",\"https://matplotlib.org\",\"jupyter.org\",\"https://github.com/llSourcell/self_driving_cars_explained?files=1\",\"https://www.youtube.com/watch?v=yt015gM-ync&feature=youtu.be\",\"https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI\",\"https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c\",\"https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8\",\"https://searchenterpriseai.techtarget.com/definition/driverless-car\",\"https://github.com/llSourcell/self_driving_cars_explained\"],\"app\":\"steemit/0.1\",\"format\":\"html\"}",
      "parent_author": "",
      "parent_permlink": "utopian-io",
      "permlink": "self-driving-cars-explained",
      "title": "Self-Driving Cars Explained"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T08:21:06",
  "trx_id": "b6e3236d659044b1cb7462fa3296021ae111e8be",
  "trx_in_block": 24,
  "virtual_op": 0
}

{
  "block": 22705496,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "<html>\n<p>http://i65.tinypic.com/2cdfkmx.jpg</p>\n<p><br></p>\n<p>This is the code for <a href=\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \">thistutorial</a> by Siraj Raval. You can find the <a href=\"https://github.com/udacity/self-driving-car-sim\">simulator here</a>.<br>\n</p>\n<h1>Overview&nbsp;</h1>\n<p>The objective of this project is to clone human driving behavior using a Deep Neural Network. In order to achieve this, we are going to use a simple Car Simulator. During the training phase, we navigate our car inside the simulator using the keyboard. While we navigating the car the simulator records training images and respective steering angles. Then we use those recorded data to train our neural network. Trained model was tested on two tracks, namely training track and validation track. Following two animations show the performance of our final model in both training and validation tracks.</p>\n<p>http://i66.tinypic.com/2a8qgxy.jpg</p>\n<h1>Github Repository:&nbsp;</h1>\n<ul>\n  <li>https://github.com/udacity/self-driving-car-sim</li>\n</ul>\n<h1>Requirements&nbsp;</h1>\n<p>This project requires <strong>Python 3.5</strong> and the following Python libraries installed:</p>\n<ul>\n  <li><a href=\"https://keras.io\">Keras</a></li>\n  <li><a href=\"www.numpy.org\">NumPy</a></li>\n  <li><a href=\"https://www.scipy.org\">SciPy</a></li>\n  <li><a href=\"https://www.tensorflow.org\">TensorFlow</a></li>\n  <li><a href=\"pandas.pydata.org\">Pandas</a></li>\n  <li><a href=\"https://opencv.org\">OpenCV</a></li>\n  <li><a href=\"https://matplotlib.org\">Matplotlib</a> (Optional)</li>\n  <li><a href=\"jupyter.org\">Jupyter</a> (Optional)</li>\n</ul>\n<p><br></p>\n<h3>All Code : https://github.com/llSourcell/self_driving_cars_explained?files=1</h3>\n<p><br></p>\n<p>Run this command at the terminal prompt to install <a href=\"https://opencv.org\">OpenCV</a>. Useful for image processing:</p>\n<ul>\n  <li><code>conda install -c&nbsp;</code></li>\n  <li><code>https://conda.anaconda.org/menpo</code>&nbsp;</li>\n  <li><code>opencv3</code></li>\n</ul>\n<h3>How to Run the Model</h3>\n<p>This repository comes with trained model which you can directly test using the following command.</p>\n<ul>\n  <li><code>python drive.py model.json</code></li>\n</ul>\n<h2>Implementation</h2>\n<h3>If you prefer watching a video..</h3>\n<p>https://www.youtube.com/watch?v=yt015gM-ync&amp;feature=youtu.be</p>\n<h2>Results</h2>\n<p>In the initial stage of the project, I used a dataset generated by myself. That dataset was small and recorded while navigating the car using the laptop keyboard. However, the model built using that dataset was not good enough to autonomously navigate the car in the simulator. However, later I used the dataset published by the Udacity. The model developed using that dataset (with the help of augmented data) works well on both tracks as shown in following videos.</p>\n<h3>More learning Lesson&nbsp;</h3>\n<ul>\n  <li>https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI</li>\n  <li>https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c</li>\n  <li>https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8</li>\n  <li>https://searchenterpriseai.techtarget.com/definition/driverless-car</li>\n</ul>\n<h2>My Repository :</h2>\n<ul>\n  <li>https://github.com/llSourcell/self_driving_cars_explained</li>\n</ul>\n</html>",
      "json_metadata": "{\"tags\":[\"utopian-io\",\"video-tutorials\",\"development\",\"technology\"],\"image\":[\"http://i65.tinypic.com/2cdfkmx.jpg\",\"http://i66.tinypic.com/2a8qgxy.jpg\",\"https://img.youtube.com/vi/yt015gM-ync/0.jpg\"],\"links\":[\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \",\"https://github.com/udacity/self-driving-car-sim\",\"https://keras.io\",\"www.numpy.org\",\"https://www.scipy.org\",\"https://www.tensorflow.org\",\"pandas.pydata.org\",\"https://opencv.org\",\"https://matplotlib.org\",\"jupyter.org\",\"https://github.com/llSourcell/self_driving_cars_explained?files=1\",\"https://www.youtube.com/watch?v=yt015gM-ync&feature=youtu.be\",\"https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI\",\"https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c\",\"https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8\",\"https://searchenterpriseai.techtarget.com/definition/driverless-car\",\"https://github.com/llSourcell/self_driving_cars_explained\"],\"app\":\"steemit/0.1\",\"format\":\"html\"}",
      "parent_author": "",
      "parent_permlink": "utopian-io",
      "permlink": "self-driving-cars-explained",
      "title": "Self-Driving Cars Explained"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T08:02:09",
  "trx_id": "d0d5717dd03aad4e9db5afe895b3d36d3cfd3057",
  "trx_in_block": 30,
  "virtual_op": 0
}

{
  "block": 22705385,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "self-driving-cars-explained",
      "voter": "anomaly",
      "weight": 100
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:56:36",
  "trx_id": "9247a198b128f0b727057b373d2252a598a87a7e",
  "trx_in_block": 40,
  "virtual_op": 0
}

{
  "block": 22705090,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "<html>\n<p>http://i65.tinypic.com/2cdfkmx.jpg</p>\n<p><br></p>\n<p>This is the code for <a href=\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \">thistutorial</a> by Siraj Raval. You can find the <a href=\"https://github.com/udacity/self-driving-car-sim\">simulator here</a>.<br>\n</p>\n<h1>Overview&nbsp;</h1>\n<p>The objective of this project is to clone human driving behavior using a Deep Neural Network. In order to achieve this, we are going to use a simple Car Simulator. During the training phase, we navigate our car inside the simulator using the keyboard. While we navigating the car the simulator records training images and respective steering angles. Then we use those recorded data to train our neural network. Trained model was tested on two tracks, namely training track and validation track. Following two animations show the performance of our final model in both training and validation tracks.</p>\n<p>http://i66.tinypic.com/2a8qgxy.jpg</p>\n<h1>Github Repository:&nbsp;</h1>\n<ul>\n  <li>https://github.com/udacity/self-driving-car-sim</li>\n</ul>\n<h1>Requirements&nbsp;</h1>\n<p>This project requires <strong>Python 3.5</strong> and the following Python libraries installed:</p>\n<ul>\n  <li><a href=\"https://keras.io\">Keras</a></li>\n  <li><a href=\"www.numpy.org\">NumPy</a></li>\n  <li><a href=\"https://www.scipy.org\">SciPy</a></li>\n  <li><a href=\"https://www.tensorflow.org\">TensorFlow</a></li>\n  <li><a href=\"pandas.pydata.org\">Pandas</a></li>\n  <li><a href=\"https://opencv.org\">OpenCV</a></li>\n  <li><a href=\"https://matplotlib.org\">Matplotlib</a> (Optional)</li>\n  <li><a href=\"jupyter.org\">Jupyter</a> (Optional)</li>\n</ul>\n<p><br></p>\n<h3>All Code : https://github.com/llSourcell/self_driving_cars_explained?files=1</h3>\n<h3>If you prefer watching a video..</h3>\n<p>https://www.youtube.com/watch?v=yt015gM-ync&amp;feature=youtu.be</p>\n<p><br></p>\n<p>Run this command at the terminal prompt to install <a href=\"https://opencv.org\">OpenCV</a>. Useful for image processing:</p>\n<ul>\n  <li><code>conda install -c&nbsp;</code></li>\n  <li><code>https://conda.anaconda.org/menpo</code>&nbsp;</li>\n  <li><code>opencv3</code></li>\n</ul>\n<h3>How to Run the Model</h3>\n<p>This repository comes with trained model which you can directly test using the following command.</p>\n<ul>\n  <li><code>python drive.py model.json</code></li>\n</ul>\n<h2>Implementation</h2>\n<h3>Data Capturing</h3>\n<p>During the training, the simulator captures data with a frequency of 10hz. Also, at a given time step it recorded three images taken from left, center, and right cameras. The following figure shows an example I have collected during the training time.</p>\n<p><br>\nhttp://i68.tinypic.com/26063og.jpg</p>\n<p>Collected data are processed before feeding into the deep neural network and those preprocessing steps are described in the latter part of this file.</p>\n<h3>Dataset Statistics</h3>\n<p>The dataset consists of 24108 images (8036 images per camera angle). The training track contains a lot of shallow turns and straight road segments. Hence, the majority of the recorded steering angles are zeros. Therefore, preprocessing images and respective steering angles are necessary in order to generalize the training model for unseen tracks such as our validation track.Next, we are going explain our data processing pipeline.</p>\n<h3>Data Processing Pipeline</h3>\n<p>The following figure shows our data preprocessing pipeline.</p>\n<p>http://i65.tinypic.com/29uwqoo.png</p>\n<p>In the very first state of the pipeline, we apply random shear operation. However, we select images with 0.9 probability for the random shearing process. We kept 10 percent of original images and steering angles in order to help the car to navigate in the training track. The following figure shows the result of shearing operation applied to a sample image.</p>\n<p>http://i63.tinypic.com/2wgylom.png</p>\n<p>The images captured by the simulator come with a lot of details which do not directly help model building process. In addition to that extra space occupied by these details required additional processing power. Hence, we remove 35 percent of the original image from the top and 10 percent. This process was done in crop stage. The following figure shows the result of cropping operation applied to an image.</p>\n<p>http://i65.tinypic.com/md38dc.png</p>\n<p>The next stage of the data processing pipeline is called random flip stage. In this stage we randomly (with 0.5 probability) flip images. The idea behind this operation is left turning bends are more prevalent than right bends in the training track. Hence, in order to increase the generalization of our mode, we flip images and respective steering angles. The following figure shows the result of flipping operation applied to an image.</p>\n<p>http://i64.tinypic.com/2mh7dli.png</p>\n<p>In the final state of the pipeline, we resize images to 64x64 in order to reduce training time. A sample resized image is shown in the following figure. Resized images are fed into the neural network. The following figure shows the result of resize operation applied to an image.</p>\n<p>http://i64.tinypic.com/2ebu6pv.png</p>\n<p>Next we are going to discuss our neural network architecture.</p>\n<h3>Network Architecture</h3>\n<p>Our convolutional neural network architecture was inspired by NVIDIA's End to End Learning for Self-Driving Cars paper. The main difference between our model and the NVIDIA mode is than we did use MaxPooling layers just after each Convolutional Layer in order to cut down training time. For more details about our network architecture please refer following figure.</p>\n<p>http://i66.tinypic.com/27wso7c.png</p>\n<p><br></p>\n<p>Training</p>\n<p>Even after cropping and resizing training images (with all augmented images), training dataset was very large and it could not fit into the main memory. Hence, we used <code>fit_generator</code> API of the Keras library for training our model.We created two generators namely:</p>\n<ul>\n  <li><code>train_gen = helper.generate_next_batch()</code></li>\n  <li><code>validation_gen = helper.generate_next_batch()</code></li>\n</ul>\n<p>Batch size of both <code>train_gen</code> and <code>validation_gen</code> was 64. We used 20032 images per training epoch. It is to be noted that these images are generated on the fly using the document processing pipeline described above. In addition to that, we used 6400 images (also generated on the fly) for validation. We used <code>Adam</code> optimizer with <code>1e-4</code> learning rate. Finally, when it comes to the number of training epochs we tried several possibilities such as <code>5</code>, <code>8</code>, <code>1</code>0, <code>2</code>5 and <code>50</code>. However, <code>8</code> works well on both training and validation tracks.</p>\n<h2>Results</h2>\n<p>In the initial stage of the project, I used a dataset generated by myself. That dataset was small and recorded while navigating the car using the laptop keyboard. However, the model built using that dataset was not good enough to autonomously navigate the car in the simulator. However, later I used the dataset published by the Udacity. The model developed using that dataset (with the help of augmented data) works well on both tracks as shown in following videos.</p>\n<h4>Training Track</h4>\n<p>http://i66.tinypic.com/6eg7s8.jpg</p>\n<h3>Validation Track</h3>\n<p>http://i67.tinypic.com/5txb0k.jpg</p>\n<p><br></p>\n<h2>Conclusions and Future Directions</h2>\n<p>In this project, we were working on a regression problem in the context of self-driving cars. In the initial phase, we mainly focused on finding a suitable network architecture and trained a model using our own dataset. According to Mean Square Error (<strong>MSE</strong>) our model worked well. However, it didn't perform as expected when we test the model using the simulator. So it was a clear indication that MSE is not a good metrics to assess the performance this project.In the next phase of the project, we started to use a new dataset (actually, it was the dataset published by Udacity). Additionally, we didn't fully rely on MSE when building our final model. Also, we use relatively small number of training epochs (namely <code>8</code> epochs). Data augmentation and new dataset work surprisingly well and our final model showed superb performance on both tracks.When it comes to extensions and future directions, I would like to highlight followings.</p>\n<ul>\n  <li>Train a model in real road conditions. For this, we might need to find a new simulator.</li>\n  <li>Experiment with other possible data augmentation techniques.</li>\n  <li>When we are driving a car, our actions such as changing steering angles and applying brakes are not just based on instantaneous driving decisions. In fact, curent driving decision is based on what was traffic/road condition in fast few seconds. Hence, it would be really interesting to seee how Recurrent Neural Network (<strong>RNN</strong>) model such as <strong>LSTM</strong> and <strong>GRU</strong> perform this problem.</li>\n  <li>Finally, training a (deep) reinforcement agent would also be an interesting additional project.</li>\n</ul>\n<p><br></p>\n<h2>More learning Lesson&nbsp;</h2>\n<ul>\n  <li>https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI</li>\n  <li>https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c</li>\n  <li>https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8</li>\n  <li>https://searchenterpriseai.techtarget.com/definition/driverless-car</li>\n</ul>\n<h2>My Repository :</h2>\n<ul>\n  <li>https://github.com/llSourcell/self_driving_cars_explained</li>\n</ul>\n<p><br></p>\n</html>",
      "json_metadata": "{\"tags\":[\"science\",\"technology\",\"video-tutorials\",\"development\",\"utopian-io\"],\"image\":[\"http://i65.tinypic.com/2cdfkmx.jpg\",\"http://i66.tinypic.com/2a8qgxy.jpg\",\"https://img.youtube.com/vi/yt015gM-ync/0.jpg\",\"http://i68.tinypic.com/26063og.jpg\",\"http://i65.tinypic.com/29uwqoo.png\",\"http://i63.tinypic.com/2wgylom.png\",\"http://i65.tinypic.com/md38dc.png\",\"http://i64.tinypic.com/2mh7dli.png\",\"http://i64.tinypic.com/2ebu6pv.png\",\"http://i66.tinypic.com/27wso7c.png\",\"http://i66.tinypic.com/6eg7s8.jpg\",\"http://i67.tinypic.com/5txb0k.jpg\"],\"links\":[\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \",\"https://github.com/udacity/self-driving-car-sim\",\"https://keras.io\",\"www.numpy.org\",\"https://www.scipy.org\",\"https://www.tensorflow.org\",\"pandas.pydata.org\",\"https://opencv.org\",\"https://matplotlib.org\",\"jupyter.org\",\"https://github.com/llSourcell/self_driving_cars_explained?files=1\",\"https://www.youtube.com/watch?v=yt015gM-ync&feature=youtu.be\",\"https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI\",\"https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c\",\"https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8\",\"https://searchenterpriseai.techtarget.com/definition/driverless-car\",\"https://github.com/llSourcell/self_driving_cars_explained\"],\"app\":\"steemit/0.1\",\"format\":\"html\"}",
      "parent_author": "",
      "parent_permlink": "utopian-io",
      "permlink": "self-driving-cars-explained",
      "title": "Self-Driving Cars Explained"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:41:51",
  "trx_id": "8a679a9e85fd36d58d171471216376935ba69778",
  "trx_in_block": 4,
  "virtual_op": 0
}

{
  "block": 22705041,
  "op": [
    "transfer",
    {
      "amount": "0.001 SBD",
      "from": "msp-reg",
      "memo": "Successful registration. Welcome to MSP & PALnet.",
      "to": "llsourcell"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:39:24",
  "trx_id": "eee6ad716242f8dfc72b328788d66e09b75024dd",
  "trx_in_block": 32,
  "virtual_op": 0
}

{
  "block": 22705029,
  "op": [
    "transfer",
    {
      "amount": "0.001 SBD",
      "from": "llsourcell",
      "memo": "dwrhz-itmjy-sqgpn",
      "to": "msp-reg"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:38:48",
  "trx_id": "3f7566eda90e1c283912f789a4d493f5ba0ee256",
  "trx_in_block": 66,
  "virtual_op": 0
}

{
  "block": 22704982,
  "op": [
    "transfer",
    {
      "amount": "0.001 SBD",
      "from": "msp-reg",
      "memo": "No valid MSP/PALnet registration found for user: llsourcell. Please check out https://minnowsupportproject.org/ for more info.",
      "to": "llsourcell"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:36:27",
  "trx_id": "a75dee82d466d5aa3d4f51f4f33a5d79470fae2d",
  "trx_in_block": 12,
  "virtual_op": 0
}

{
  "block": 22704980,
  "op": [
    "transfer",
    {
      "amount": "0.001 SBD",
      "from": "llsourcell",
      "memo": "yqnsz-vdhxm-jtgof",
      "to": "msp-reg"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:36:21",
  "trx_id": "de5542fa6fb84c99fa214be62cfb63299f5091fc",
  "trx_in_block": 6,
  "virtual_op": 0
}

{
  "block": 22704875,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "<html>\n<p>http://i65.tinypic.com/2cdfkmx.jpg</p>\n<p><br></p>\n<p>This is the code for <a href=\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \">thistutorial</a> by Siraj Raval. You can find the <a href=\"https://github.com/udacity/self-driving-car-sim\">simulator here</a>.<br>\n</p>\n<h1>Overview&nbsp;</h1>\n<p>The objective of this project is to clone human driving behavior using a Deep Neural Network. In order to achieve this, we are going to use a simple Car Simulator. During the training phase, we navigate our car inside the simulator using the keyboard. While we navigating the car the simulator records training images and respective steering angles. Then we use those recorded data to train our neural network. Trained model was tested on two tracks, namely training track and validation track. Following two animations show the performance of our final model in both training and validation tracks.</p>\n<p>http://i66.tinypic.com/2a8qgxy.jpg</p>\n<h1>Github Repository:&nbsp;</h1>\n<ul>\n  <li>https://github.com/udacity/self-driving-car-sim</li>\n</ul>\n<h1>Requirements&nbsp;</h1>\n<p>This project requires <strong>Python 3.5</strong> and the following Python libraries installed:</p>\n<ul>\n  <li><a href=\"https://keras.io\">Keras</a></li>\n  <li><a href=\"www.numpy.org\">NumPy</a></li>\n  <li><a href=\"https://www.scipy.org\">SciPy</a></li>\n  <li><a href=\"https://www.tensorflow.org\">TensorFlow</a></li>\n  <li><a href=\"pandas.pydata.org\">Pandas</a></li>\n  <li><a href=\"https://opencv.org\">OpenCV</a></li>\n  <li><a href=\"https://matplotlib.org\">Matplotlib</a> (Optional)</li>\n  <li><a href=\"jupyter.org\">Jupyter</a> (Optional)</li>\n</ul>\n<p><br></p>\n<h3>All Code : https://github.com/llSourcell/self_driving_cars_explained?files=1</h3>\n<h3>If you prefer watching a video..</h3>\n<p>https://www.youtube.com/watch?v=yt015gM-ync&amp;feature=youtu.be</p>\n<p><br></p>\n<p>Run this command at the terminal prompt to install <a href=\"https://opencv.org\">OpenCV</a>. Useful for image processing:</p>\n<ul>\n  <li><code>conda install -c&nbsp;</code></li>\n  <li><code>https://conda.anaconda.org/menpo</code>&nbsp;</li>\n  <li><code>opencv3</code></li>\n</ul>\n<h3>How to Run the Model</h3>\n<p>This repository comes with trained model which you can directly test using the following command.</p>\n<ul>\n  <li><code>python drive.py model.json</code></li>\n</ul>\n<h2>Implementation</h2>\n<h3>Data Capturing</h3>\n<p>During the training, the simulator captures data with a frequency of 10hz. Also, at a given time step it recorded three images taken from left, center, and right cameras. The following figure shows an example I have collected during the training time.</p>\n<p><br>\nhttp://i68.tinypic.com/26063og.jpg</p>\n<p>Collected data are processed before feeding into the deep neural network and those preprocessing steps are described in the latter part of this file.</p>\n<h3>Dataset Statistics</h3>\n<p>The dataset consists of 24108 images (8036 images per camera angle). The training track contains a lot of shallow turns and straight road segments. Hence, the majority of the recorded steering angles are zeros. Therefore, preprocessing images and respective steering angles are necessary in order to generalize the training model for unseen tracks such as our validation track.Next, we are going explain our data processing pipeline.</p>\n<h3>Data Processing Pipeline</h3>\n<p>The following figure shows our data preprocessing pipeline.</p>\n<p>http://i65.tinypic.com/29uwqoo.png</p>\n<p>In the very first state of the pipeline, we apply random shear operation. However, we select images with 0.9 probability for the random shearing process. We kept 10 percent of original images and steering angles in order to help the car to navigate in the training track. The following figure shows the result of shearing operation applied to a sample image.</p>\n<p>http://i63.tinypic.com/2wgylom.png</p>\n<p>The images captured by the simulator come with a lot of details which do not directly help model building process. In addition to that extra space occupied by these details required additional processing power. Hence, we remove 35 percent of the original image from the top and 10 percent. This process was done in crop stage. The following figure shows the result of cropping operation applied to an image.</p>\n<p>http://i65.tinypic.com/md38dc.png</p>\n<p>The next stage of the data processing pipeline is called random flip stage. In this stage we randomly (with 0.5 probability) flip images. The idea behind this operation is left turning bends are more prevalent than right bends in the training track. Hence, in order to increase the generalization of our mode, we flip images and respective steering angles. The following figure shows the result of flipping operation applied to an image.</p>\n<p>http://i64.tinypic.com/2mh7dli.png</p>\n<p>In the final state of the pipeline, we resize images to 64x64 in order to reduce training time. A sample resized image is shown in the following figure. Resized images are fed into the neural network. The following figure shows the result of resize operation applied to an image.</p>\n<p>http://i64.tinypic.com/2ebu6pv.png</p>\n<p>Next we are going to discuss our neural network architecture.</p>\n<h3>Network Architecture</h3>\n<p>Our convolutional neural network architecture was inspired by NVIDIA's End to End Learning for Self-Driving Cars paper. The main difference between our model and the NVIDIA mode is than we did use MaxPooling layers just after each Convolutional Layer in order to cut down training time. For more details about our network architecture please refer following figure.</p>\n<p>http://i66.tinypic.com/27wso7c.png</p>\n<p><br></p>\n<p>Training</p>\n<p>Even after cropping and resizing training images (with all augmented images), training dataset was very large and it could not fit into the main memory. Hence, we used <code>fit_generator</code> API of the Keras library for training our model.We created two generators namely:</p>\n<ul>\n  <li><code>train_gen = helper.generate_next_batch()</code></li>\n  <li><code>validation_gen = helper.generate_next_batch()</code></li>\n</ul>\n<p>Batch size of both <code>train_gen</code> and <code>validation_gen</code> was 64. We used 20032 images per training epoch. It is to be noted that these images are generated on the fly using the document processing pipeline described above. In addition to that, we used 6400 images (also generated on the fly) for validation. We used <code>Adam</code> optimizer with <code>1e-4</code> learning rate. Finally, when it comes to the number of training epochs we tried several possibilities such as <code>5</code>, <code>8</code>, <code>1</code>0, <code>2</code>5 and <code>50</code>. However, <code>8</code> works well on both training and validation tracks.</p>\n<h2>Results</h2>\n<p>In the initial stage of the project, I used a dataset generated by myself. That dataset was small and recorded while navigating the car using the laptop keyboard. However, the model built using that dataset was not good enough to autonomously navigate the car in the simulator. However, later I used the dataset published by the Udacity. The model developed using that dataset (with the help of augmented data) works well on both tracks as shown in following videos.</p>\n<h4>Training Track</h4>\n<p>http://i66.tinypic.com/6eg7s8.jpg</p>\n<h3>Validation Track</h3>\n<p>http://i67.tinypic.com/5txb0k.jpg</p>\n<p><br></p>\n<h2>Conclusions and Future Directions</h2>\n<p>In this project, we were working on a regression problem in the context of self-driving cars. In the initial phase, we mainly focused on finding a suitable network architecture and trained a model using our own dataset. According to Mean Square Error (<strong>MSE</strong>) our model worked well. However, it didn't perform as expected when we test the model using the simulator. So it was a clear indication that MSE is not a good metrics to assess the performance this project.In the next phase of the project, we started to use a new dataset (actually, it was the dataset published by Udacity). Additionally, we didn't fully rely on MSE when building our final model. Also, we use relatively small number of training epochs (namely <code>8</code> epochs). Data augmentation and new dataset work surprisingly well and our final model showed superb performance on both tracks.When it comes to extensions and future directions, I would like to highlight followings.</p>\n<ul>\n  <li>Train a model in real road conditions. For this, we might need to find a new simulator.</li>\n  <li>Experiment with other possible data augmentation techniques.</li>\n  <li>When we are driving a car, our actions such as changing steering angles and applying brakes are not just based on instantaneous driving decisions. In fact, curent driving decision is based on what was traffic/road condition in fast few seconds. Hence, it would be really interesting to seee how Recurrent Neural Network (<strong>RNN</strong>) model such as <strong>LSTM</strong> and <strong>GRU</strong> perform this problem.</li>\n  <li>Finally, training a (deep) reinforcement agent would also be an interesting additional project.</li>\n</ul>\n<p><br></p>\n<h2>More learning Lesson&nbsp;</h2>\n<ul>\n  <li>https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI</li>\n  <li>https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c</li>\n  <li>https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8</li>\n  <li>https://searchenterpriseai.techtarget.com/definition/driverless-car</li>\n</ul>\n<h2>My Repository :</h2>\n<ul>\n  <li>https://github.com/llSourcell/self_driving_cars_explained</li>\n</ul>\n<p><br></p>\n</html>",
      "json_metadata": "{\"tags\":[\"technology\",\"video-tutorials\",\"development\",\"utopian-io\"],\"image\":[\"http://i65.tinypic.com/2cdfkmx.jpg\",\"http://i66.tinypic.com/2a8qgxy.jpg\",\"https://img.youtube.com/vi/yt015gM-ync/0.jpg\",\"http://i68.tinypic.com/26063og.jpg\",\"http://i65.tinypic.com/29uwqoo.png\",\"http://i63.tinypic.com/2wgylom.png\",\"http://i65.tinypic.com/md38dc.png\",\"http://i64.tinypic.com/2mh7dli.png\",\"http://i64.tinypic.com/2ebu6pv.png\",\"http://i66.tinypic.com/27wso7c.png\",\"http://i66.tinypic.com/6eg7s8.jpg\",\"http://i67.tinypic.com/5txb0k.jpg\"],\"links\":[\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \",\"https://github.com/udacity/self-driving-car-sim\",\"https://keras.io\",\"www.numpy.org\",\"https://www.scipy.org\",\"https://www.tensorflow.org\",\"pandas.pydata.org\",\"https://opencv.org\",\"https://matplotlib.org\",\"jupyter.org\",\"https://github.com/llSourcell/self_driving_cars_explained?files=1\",\"https://www.youtube.com/watch?v=yt015gM-ync&feature=youtu.be\",\"https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI\",\"https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c\",\"https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8\",\"https://searchenterpriseai.techtarget.com/definition/driverless-car\",\"https://github.com/llSourcell/self_driving_cars_explained\"],\"app\":\"steemit/0.1\",\"format\":\"html\"}",
      "parent_author": "",
      "parent_permlink": "utopian-io",
      "permlink": "self-driving-cars-explained",
      "title": "Self-Driving Cars Explained"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:31:06",
  "trx_id": "b547be0426c7b4b44b00be83fefdb13d64959d5a",
  "trx_in_block": 35,
  "virtual_op": 0
}

{
  "block": 22704806,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "<html>\n<p>http://i65.tinypic.com/2cdfkmx.jpg</p>\n<p><br></p>\n<p>This is the code for <a href=\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \">thistutorial</a> by Siraj Raval. You can find the <a href=\"https://github.com/udacity/self-driving-car-sim\">simulator here</a>.<br>\n</p>\n<h1>Overview&nbsp;</h1>\n<p>The objective of this project is to clone human driving behavior using a Deep Neural Network. In order to achieve this, we are going to use a simple Car Simulator. During the training phase, we navigate our car inside the simulator using the keyboard. While we navigating the car the simulator records training images and respective steering angles. Then we use those recorded data to train our neural network. Trained model was tested on two tracks, namely training track and validation track. Following two animations show the performance of our final model in both training and validation tracks.</p>\n<p>http://i66.tinypic.com/2a8qgxy.jpg</p>\n<h1>Github Repository:&nbsp;</h1>\n<ul>\n  <li>https://github.com/udacity/self-driving-car-sim</li>\n</ul>\n<h1>Requirements&nbsp;</h1>\n<p>This project requires <strong>Python 3.5</strong> and the following Python libraries installed:</p>\n<ul>\n  <li><a href=\"https://keras.io\">Keras</a></li>\n  <li><a href=\"www.numpy.org\">NumPy</a></li>\n  <li><a href=\"https://www.scipy.org\">SciPy</a></li>\n  <li><a href=\"https://www.tensorflow.org\">TensorFlow</a></li>\n  <li><a href=\"pandas.pydata.org\">Pandas</a></li>\n  <li><a href=\"https://opencv.org\">OpenCV</a></li>\n  <li><a href=\"https://matplotlib.org\">Matplotlib</a> (Optional)</li>\n  <li><a href=\"jupyter.org\">Jupyter</a> (Optional)</li>\n</ul>\n<p><br></p>\n<h3>All Code : https://github.com/llSourcell/self_driving_cars_explained?files=1</h3>\n<h3>If you prefer watching a video..</h3>\n<p>https://www.youtube.com/watch?v=yt015gM-ync&amp;feature=youtu.be</p>\n<p><br></p>\n<p>Run this command at the terminal prompt to install <a href=\"https://opencv.org\">OpenCV</a>. Useful for image processing:</p>\n<ul>\n  <li><code>conda install -c&nbsp;</code></li>\n  <li><code>https://conda.anaconda.org/menpo</code>&nbsp;</li>\n  <li><code>opencv3</code></li>\n</ul>\n<h3>How to Run the Model</h3>\n<p>This repository comes with trained model which you can directly test using the following command.</p>\n<ul>\n  <li><code>python drive.py model.json</code></li>\n</ul>\n<h2>Implementation</h2>\n<h3>Data Capturing</h3>\n<p>During the training, the simulator captures data with a frequency of 10hz. Also, at a given time step it recorded three images taken from left, center, and right cameras. The following figure shows an example I have collected during the training time.</p>\n<p><br>\nhttp://i68.tinypic.com/26063og.jpg</p>\n<p>Collected data are processed before feeding into the deep neural network and those preprocessing steps are described in the latter part of this file.</p>\n<h3>Dataset Statistics</h3>\n<p>The dataset consists of 24108 images (8036 images per camera angle). The training track contains a lot of shallow turns and straight road segments. Hence, the majority of the recorded steering angles are zeros. Therefore, preprocessing images and respective steering angles are necessary in order to generalize the training model for unseen tracks such as our validation track.Next, we are going explain our data processing pipeline.</p>\n<h3>Data Processing Pipeline</h3>\n<p>The following figure shows our data preprocessing pipeline.</p>\n<p>http://i65.tinypic.com/29uwqoo.png</p>\n<p>In the very first state of the pipeline, we apply random shear operation. However, we select images with 0.9 probability for the random shearing process. We kept 10 percent of original images and steering angles in order to help the car to navigate in the training track. The following figure shows the result of shearing operation applied to a sample image.</p>\n<p>http://i63.tinypic.com/2wgylom.png</p>\n<p>The images captured by the simulator come with a lot of details which do not directly help model building process. In addition to that extra space occupied by these details required additional processing power. Hence, we remove 35 percent of the original image from the top and 10 percent. This process was done in crop stage. The following figure shows the result of cropping operation applied to an image.</p>\n<p>http://i65.tinypic.com/md38dc.png</p>\n<p>The next stage of the data processing pipeline is called random flip stage. In this stage we randomly (with 0.5 probability) flip images. The idea behind this operation is left turning bends are more prevalent than right bends in the training track. Hence, in order to increase the generalization of our mode, we flip images and respective steering angles. The following figure shows the result of flipping operation applied to an image.</p>\n<p>http://i64.tinypic.com/2mh7dli.png</p>\n<p>In the final state of the pipeline, we resize images to 64x64 in order to reduce training time. A sample resized image is shown in the following figure. Resized images are fed into the neural network. The following figure shows the result of resize operation applied to an image.</p>\n<p>http://i64.tinypic.com/2ebu6pv.png</p>\n<p>Next we are going to discuss our neural network architecture.</p>\n<h3>Network Architecture</h3>\n<p>Our convolutional neural network architecture was inspired by NVIDIA's End to End Learning for Self-Driving Cars paper. The main difference between our model and the NVIDIA mode is than we did use MaxPooling layers just after each Convolutional Layer in order to cut down training time. For more details about our network architecture please refer following figure.</p>\n<p>http://i66.tinypic.com/27wso7c.png</p>\n<p><br></p>\n<p>Training</p>\n<p>Even after cropping and resizing training images (with all augmented images), training dataset was very large and it could not fit into the main memory. Hence, we used <code>fit_generator</code> API of the Keras library for training our model.We created two generators namely:</p>\n<ul>\n  <li><code>train_gen = helper.generate_next_batch()</code></li>\n  <li><code>validation_gen = helper.generate_next_batch()</code></li>\n</ul>\n<p>Batch size of both <code>train_gen</code> and <code>validation_gen</code> was 64. We used 20032 images per training epoch. It is to be noted that these images are generated on the fly using the document processing pipeline described above. In addition to that, we used 6400 images (also generated on the fly) for validation. We used <code>Adam</code> optimizer with <code>1e-4</code> learning rate. Finally, when it comes to the number of training epochs we tried several possibilities such as <code>5</code>, <code>8</code>, <code>1</code>0, <code>2</code>5 and <code>50</code>. However, <code>8</code> works well on both training and validation tracks.</p>\n<h2>Results</h2>\n<p>In the initial stage of the project, I used a dataset generated by myself. That dataset was small and recorded while navigating the car using the laptop keyboard. However, the model built using that dataset was not good enough to autonomously navigate the car in the simulator. However, later I used the dataset published by the Udacity. The model developed using that dataset (with the help of augmented data) works well on both tracks as shown in following videos.</p>\n<h4>Training Track</h4>\n<p>http://i66.tinypic.com/6eg7s8.jpg</p>\n<h3>Validation Track</h3>\n<p>http://i67.tinypic.com/5txb0k.jpg</p>\n<p><br></p>\n<h2>Conclusions and Future Directions</h2>\n<p>In this project, we were working on a regression problem in the context of self-driving cars. In the initial phase, we mainly focused on finding a suitable network architecture and trained a model using our own dataset. According to Mean Square Error (<strong>MSE</strong>) our model worked well. However, it didn't perform as expected when we test the model using the simulator. So it was a clear indication that MSE is not a good metrics to assess the performance this project.In the next phase of the project, we started to use a new dataset (actually, it was the dataset published by Udacity). Additionally, we didn't fully rely on MSE when building our final model. Also, we use relatively small number of training epochs (namely <code>8</code> epochs). Data augmentation and new dataset work surprisingly well and our final model showed superb performance on both tracks.When it comes to extensions and future directions, I would like to highlight followings.</p>\n<ul>\n  <li>Train a model in real road conditions. For this, we might need to find a new simulator.</li>\n  <li>Experiment with other possible data augmentation techniques.</li>\n  <li>When we are driving a car, our actions such as changing steering angles and applying brakes are not just based on instantaneous driving decisions. In fact, curent driving decision is based on what was traffic/road condition in fast few seconds. Hence, it would be really interesting to seee how Recurrent Neural Network (<strong>RNN</strong>) model such as <strong>LSTM</strong> and <strong>GRU</strong> perform this problem.</li>\n  <li>Finally, training a (deep) reinforcement agent would also be an interesting additional project.</li>\n</ul>\n<p><br></p>\n<h2>More learning Lesson&nbsp;</h2>\n<ul>\n  <li>https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI</li>\n  <li>https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c</li>\n  <li>https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8</li>\n  <li>https://searchenterpriseai.techtarget.com/definition/driverless-car</li>\n</ul>\n<h2>My Repository :</h2>\n<ul>\n  <li>https://github.com/llSourcell/self_driving_cars_explained</li>\n</ul>\n<p><br></p>\n</html>",
      "json_metadata": "{\"tags\":[\"tutorials\",\"development\",\"technology\",\"video-tutorials\",\"utopian-io\"],\"image\":[\"http://i65.tinypic.com/2cdfkmx.jpg\",\"http://i66.tinypic.com/2a8qgxy.jpg\",\"https://img.youtube.com/vi/yt015gM-ync/0.jpg\",\"http://i68.tinypic.com/26063og.jpg\",\"http://i65.tinypic.com/29uwqoo.png\",\"http://i63.tinypic.com/2wgylom.png\",\"http://i65.tinypic.com/md38dc.png\",\"http://i64.tinypic.com/2mh7dli.png\",\"http://i64.tinypic.com/2ebu6pv.png\",\"http://i66.tinypic.com/27wso7c.png\",\"http://i66.tinypic.com/6eg7s8.jpg\",\"http://i67.tinypic.com/5txb0k.jpg\"],\"links\":[\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \",\"https://github.com/udacity/self-driving-car-sim\",\"https://keras.io\",\"www.numpy.org\",\"https://www.scipy.org\",\"https://www.tensorflow.org\",\"pandas.pydata.org\",\"https://opencv.org\",\"https://matplotlib.org\",\"jupyter.org\",\"https://github.com/llSourcell/self_driving_cars_explained?files=1\",\"https://www.youtube.com/watch?v=yt015gM-ync&feature=youtu.be\",\"https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI\",\"https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c\",\"https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8\",\"https://searchenterpriseai.techtarget.com/definition/driverless-car\",\"https://github.com/llSourcell/self_driving_cars_explained\"],\"app\":\"steemit/0.1\",\"format\":\"html\"}",
      "parent_author": "",
      "parent_permlink": "utopian-io",
      "permlink": "self-driving-cars-explained",
      "title": "Self-Driving Cars Explained"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:27:39",
  "trx_id": "7a8f15bcf713df89411d656b8a6615ba0ffffdee",
  "trx_in_block": 20,
  "virtual_op": 0
}

{
  "block": 22704757,
  "op": [
    "comment",
    {
      "author": "cheetah",
      "body": "Hi! I am a robot. I just upvoted you! I found similar content that readers might be interested in:\nhttps://github.com/upul/Behavioral-Cloning",
      "json_metadata": "",
      "parent_author": "llsourcell",
      "parent_permlink": "self-driving-cars-explained",
      "permlink": "cheetah-re-llsourcellself-driving-cars-explained",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:25:12",
  "trx_id": "47bbf967d1396e06a7ee956b743443d4cada485a",
  "trx_in_block": 32,
  "virtual_op": 0
}

{
  "block": 22704755,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "self-driving-cars-explained",
      "voter": "cheetah",
      "weight": 8
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:25:06",
  "trx_id": "fd2f5bc1c67480536501970f12bda1b60bcd4a6f",
  "trx_in_block": 52,
  "virtual_op": 0
}

{
  "block": 22704754,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "self-driving-cars-explained",
      "voter": "ax3",
      "weight": 100
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:25:03",
  "trx_id": "b86e416a3b9cd46a81ace19fe41168482a35c8d3",
  "trx_in_block": 3,
  "virtual_op": 0
}

{
  "block": 22704748,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "<html>\n<p>http://i65.tinypic.com/2cdfkmx.jpg</p>\n<p><br></p>\n<p>This is the code for <a href=\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \">thistutorial</a> by Siraj Raval. You can find the <a href=\"https://github.com/udacity/self-driving-car-sim\">simulator here</a>.<br>\n</p>\n<h1>Overview&nbsp;</h1>\n<p>The objective of this project is to clone human driving behavior using a Deep Neural Network. In order to achieve this, we are going to use a simple Car Simulator. During the training phase, we navigate our car inside the simulator using the keyboard. While we navigating the car the simulator records training images and respective steering angles. Then we use those recorded data to train our neural network. Trained model was tested on two tracks, namely training track and validation track. Following two animations show the performance of our final model in both training and validation tracks.</p>\n<p>http://i66.tinypic.com/2a8qgxy.jpg</p>\n<h1>Github Repository:&nbsp;</h1>\n<ul>\n  <li>https://github.com/udacity/self-driving-car-sim</li>\n</ul>\n<h1>Requirements&nbsp;</h1>\n<p>This project requires <strong>Python 3.5</strong> and the following Python libraries installed:</p>\n<ul>\n  <li><a href=\"https://keras.io\">Keras</a></li>\n  <li><a href=\"www.numpy.org\">NumPy</a></li>\n  <li><a href=\"https://www.scipy.org\">SciPy</a></li>\n  <li><a href=\"https://www.tensorflow.org\">TensorFlow</a></li>\n  <li><a href=\"pandas.pydata.org\">Pandas</a></li>\n  <li><a href=\"https://opencv.org\">OpenCV</a></li>\n  <li><a href=\"https://matplotlib.org\">Matplotlib</a> (Optional)</li>\n  <li><a href=\"jupyter.org\">Jupyter</a> (Optional)</li>\n</ul>\n<p><br></p>\n<h3>All Code : https://github.com/llSourcell/self_driving_cars_explained?files=1</h3>\n<h3>If you prefer watching a video..</h3>\n<p>https://www.youtube.com/watch?v=yt015gM-ync&amp;feature=youtu.be</p>\n<p><br></p>\n<p>Run this command at the terminal prompt to install <a href=\"https://opencv.org\">OpenCV</a>. Useful for image processing:</p>\n<ul>\n  <li><code>conda install -c&nbsp;</code></li>\n  <li><code>https://conda.anaconda.org/menpo</code>&nbsp;</li>\n  <li><code>opencv3</code></li>\n</ul>\n<h3>How to Run the Model</h3>\n<p>This repository comes with trained model which you can directly test using the following command.</p>\n<ul>\n  <li><code>python drive.py model.json</code></li>\n</ul>\n<h2>Implementation</h2>\n<h3>Data Capturing</h3>\n<p>During the training, the simulator captures data with a frequency of 10hz. Also, at a given time step it recorded three images taken from left, center, and right cameras. The following figure shows an example I have collected during the training time.</p>\n<p><br>\nhttp://i68.tinypic.com/26063og.jpg</p>\n<p>Collected data are processed before feeding into the deep neural network and those preprocessing steps are described in the latter part of this file.</p>\n<h3>Dataset Statistics</h3>\n<p>The dataset consists of 24108 images (8036 images per camera angle). The training track contains a lot of shallow turns and straight road segments. Hence, the majority of the recorded steering angles are zeros. Therefore, preprocessing images and respective steering angles are necessary in order to generalize the training model for unseen tracks such as our validation track.Next, we are going explain our data processing pipeline.</p>\n<h3>Data Processing Pipeline</h3>\n<p>The following figure shows our data preprocessing pipeline.</p>\n<p>http://i65.tinypic.com/29uwqoo.png</p>\n<p>In the very first state of the pipeline, we apply random shear operation. However, we select images with 0.9 probability for the random shearing process. We kept 10 percent of original images and steering angles in order to help the car to navigate in the training track. The following figure shows the result of shearing operation applied to a sample image.</p>\n<p>http://i63.tinypic.com/2wgylom.png</p>\n<p>The images captured by the simulator come with a lot of details which do not directly help model building process. In addition to that extra space occupied by these details required additional processing power. Hence, we remove 35 percent of the original image from the top and 10 percent. This process was done in crop stage. The following figure shows the result of cropping operation applied to an image.</p>\n<p>http://i65.tinypic.com/md38dc.png</p>\n<p>The next stage of the data processing pipeline is called random flip stage. In this stage we randomly (with 0.5 probability) flip images. The idea behind this operation is left turning bends are more prevalent than right bends in the training track. Hence, in order to increase the generalization of our mode, we flip images and respective steering angles. The following figure shows the result of flipping operation applied to an image.</p>\n<p>http://i64.tinypic.com/2mh7dli.png</p>\n<p>In the final state of the pipeline, we resize images to 64x64 in order to reduce training time. A sample resized image is shown in the following figure. Resized images are fed into the neural network. The following figure shows the result of resize operation applied to an image.</p>\n<p>http://i64.tinypic.com/2ebu6pv.png</p>\n<p>Next we are going to discuss our neural network architecture.</p>\n<h3>Network Architecture</h3>\n<p>Our convolutional neural network architecture was inspired by NVIDIA's End to End Learning for Self-Driving Cars paper. The main difference between our model and the NVIDIA mode is than we did use MaxPooling layers just after each Convolutional Layer in order to cut down training time. For more details about our network architecture please refer following figure.</p>\n<p>http://i66.tinypic.com/27wso7c.png</p>\n<p><br></p>\n<p>Training</p>\n<p>Even after cropping and resizing training images (with all augmented images), training dataset was very large and it could not fit into the main memory. Hence, we used <code>fit_generator</code> API of the Keras library for training our model.We created two generators namely:</p>\n<ul>\n  <li><code>train_gen = helper.generate_next_batch()</code></li>\n  <li><code>validation_gen = helper.generate_next_batch()</code></li>\n</ul>\n<p>Batch size of both <code>train_gen</code> and <code>validation_gen</code> was 64. We used 20032 images per training epoch. It is to be noted that these images are generated on the fly using the document processing pipeline described above. In addition to that, we used 6400 images (also generated on the fly) for validation. We used <code>Adam</code> optimizer with <code>1e-4</code> learning rate. Finally, when it comes to the number of training epochs we tried several possibilities such as <code>5</code>, <code>8</code>, <code>1</code>0, <code>2</code>5 and <code>50</code>. However, <code>8</code> works well on both training and validation tracks.</p>\n<h2>Results</h2>\n<p>In the initial stage of the project, I used a dataset generated by myself. That dataset was small and recorded while navigating the car using the laptop keyboard. However, the model built using that dataset was not good enough to autonomously navigate the car in the simulator. However, later I used the dataset published by the Udacity. The model developed using that dataset (with the help of augmented data) works well on both tracks as shown in following videos.</p>\n<h4>Training Track</h4>\n<p>http://i66.tinypic.com/6eg7s8.jpg</p>\n<h3>Validation Track</h3>\n<p>http://i67.tinypic.com/5txb0k.jpg</p>\n<p><br></p>\n<h2>Conclusions and Future Directions</h2>\n<p>In this project, we were working on a regression problem in the context of self-driving cars. In the initial phase, we mainly focused on finding a suitable network architecture and trained a model using our own dataset. According to Mean Square Error (<strong>MSE</strong>) our model worked well. However, it didn't perform as expected when we test the model using the simulator. So it was a clear indication that MSE is not a good metrics to assess the performance this project.In the next phase of the project, we started to use a new dataset (actually, it was the dataset published by Udacity). Additionally, we didn't fully rely on MSE when building our final model. Also, we use relatively small number of training epochs (namely <code>8</code> epochs). Data augmentation and new dataset work surprisingly well and our final model showed superb performance on both tracks.When it comes to extensions and future directions, I would like to highlight followings.</p>\n<ul>\n  <li>Train a model in real road conditions. For this, we might need to find a new simulator.</li>\n  <li>Experiment with other possible data augmentation techniques.</li>\n  <li>When we are driving a car, our actions such as changing steering angles and applying brakes are not just based on instantaneous driving decisions. In fact, curent driving decision is based on what was traffic/road condition in fast few seconds. Hence, it would be really interesting to seee how Recurrent Neural Network (<strong>RNN</strong>) model such as <strong>LSTM</strong> and <strong>GRU</strong> perform this problem.</li>\n  <li>Finally, training a (deep) reinforcement agent would also be an interesting additional project.</li>\n</ul>\n<p><br></p>\n<h2>More learning Lesson&nbsp;</h2>\n<ul>\n  <li>https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI</li>\n  <li>https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c</li>\n  <li>https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8</li>\n  <li>https://searchenterpriseai.techtarget.com/definition/driverless-car</li>\n</ul>\n<h2>My Repository :</h2>\n<ul>\n  <li>https://github.com/llSourcell/self_driving_cars_explained</li>\n</ul>\n<p><br></p>\n</html>",
      "json_metadata": "{\"tags\":[\"utopian-io\",\"tutorials\",\"development\",\"technology\",\"video-tutorials\"],\"image\":[\"http://i65.tinypic.com/2cdfkmx.jpg\",\"http://i66.tinypic.com/2a8qgxy.jpg\",\"https://img.youtube.com/vi/yt015gM-ync/0.jpg\",\"http://i68.tinypic.com/26063og.jpg\",\"http://i65.tinypic.com/29uwqoo.png\",\"http://i63.tinypic.com/2wgylom.png\",\"http://i65.tinypic.com/md38dc.png\",\"http://i64.tinypic.com/2mh7dli.png\",\"http://i64.tinypic.com/2ebu6pv.png\",\"http://i66.tinypic.com/27wso7c.png\",\"http://i66.tinypic.com/6eg7s8.jpg\",\"http://i67.tinypic.com/5txb0k.jpg\"],\"links\":[\"https://github.com/llSourcell/self_driving_cars_explained?files=1 \",\"https://github.com/udacity/self-driving-car-sim\",\"https://keras.io\",\"www.numpy.org\",\"https://www.scipy.org\",\"https://www.tensorflow.org\",\"pandas.pydata.org\",\"https://opencv.org\",\"https://matplotlib.org\",\"jupyter.org\",\"https://github.com/llSourcell/self_driving_cars_explained?files=1\",\"https://www.youtube.com/watch?v=yt015gM-ync&feature=youtu.be\",\"https://www.ucsusa.org/clean-vehicles/how-self-driving-cars-work#.WwGlx9MvwmI\",\"https://medium.com/swlh/everything-about-self-driving-cars-explained-for-non-engineers-f73997dcb60c\",\"https://hackernoon.com/self-driving-cars-explained-db9fc8ced7e8\",\"https://searchenterpriseai.techtarget.com/definition/driverless-car\",\"https://github.com/llSourcell/self_driving_cars_explained\"],\"app\":\"steemit/0.1\",\"format\":\"html\"}",
      "parent_author": "",
      "parent_permlink": "utopian-io",
      "permlink": "self-driving-cars-explained",
      "title": "Self-Driving Cars Explained"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T07:24:45",
  "trx_id": "9a0a5b8516bc786b6007cc66a6f37c37854e2a4a",
  "trx_in_block": 1,
  "virtual_op": 0
}

{
  "block": 22702977,
  "op": [
    "comment",
    {
      "author": "joeyarnoldvn",
      "body": "You are fun.",
      "json_metadata": "{\"tags\":[\"introduceyourself\"],\"app\":\"steemit/0.1\"}",
      "parent_author": "llsourcell",
      "parent_permlink": "introducemyself-welcome-steemit-folks",
      "permlink": "re-llsourcell-introducemyself-welcome-steemit-folks-20180524t055611982z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T05:56:12",
  "trx_id": "8166f586954df853fc5c0d69484ee7b4b073a7ee",
  "trx_in_block": 23,
  "virtual_op": 0
}

{
  "block": 22701611,
  "op": [
    "vote",
    {
      "author": "fyrstikken",
      "permlink": "if-you-trail-this-secret-account-you-make-more-money-as-curator",
      "voter": "llsourcell",
      "weight": 10000
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:47:54",
  "trx_id": "ea7e60b1473df57c652c724f45d28fceecd08905",
  "trx_in_block": 15,
  "virtual_op": 0
}

{
  "block": 22701607,
  "op": [
    "vote",
    {
      "author": "blocktrades",
      "permlink": "ann-blocktrades-is-now-buying-selling-monero",
      "voter": "llsourcell",
      "weight": 10000
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:47:42",
  "trx_id": "d51398456d7fb6cdeb069ae8f64f73401691cfe3",
  "trx_in_block": 15,
  "virtual_op": 0
}

{
  "block": 22701599,
  "op": [
    "vote",
    {
      "author": "adonisabril",
      "permlink": "sunset-over-seattle",
      "voter": "llsourcell",
      "weight": 10000
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:47:18",
  "trx_id": "d13d96dea9f183753351a09a48ddc121f5acb50e",
  "trx_in_block": 30,
  "virtual_op": 0
}

{
  "block": 22701592,
  "op": [
    "vote",
    {
      "author": "happymoneyman",
      "permlink": "eos-nears-launch-a-talk-with-thomas-cox",
      "voter": "llsourcell",
      "weight": 10000
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:46:57",
  "trx_id": "70620947c197bc1d37a19289c63f9d79a50d1efc",
  "trx_in_block": 53,
  "virtual_op": 0
}

{
  "block": 22701421,
  "op": [
    "account_update",
    {
      "account": "llsourcell",
      "json_metadata": "{\"profile\":{\"profile_image\":\"http://i66.tinypic.com/s4vq0y.png\",\"name\":\"Siraj Raval\",\"about\":\"Director at School of AI, inspire and educate developers to build AI.\",\"location\":\"San Francisco, CA\",\"website\":\"https://www.youtube.com/c/sirajraval\"}}",
      "memo_key": "STM7tn6mUDrHTLfT86EPoNGAM36NVSRx7hy7fKBDApRv1pwrMaXu8"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:38:24",
  "trx_id": "b054d2d29accb8868fdd4df52c667364bec9b8d8",
  "trx_in_block": 56,
  "virtual_op": 0
}

{
  "block": 22701240,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "introducemyself-welcome-steemit-folks",
      "voter": "dlivestarbooster",
      "weight": 200
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:29:21",
  "trx_id": "0c5c606ba3c816cf3e402e1f3d8d8b6baa4a90c4",
  "trx_in_block": 48,
  "virtual_op": 0
}

{
  "block": 22701239,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "introducemyself-welcome-steemit-folks",
      "voter": "kingkong1",
      "weight": 200
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:29:18",
  "trx_id": "c42bb56b5284d965b1aaad9d2e103e8de5423c89",
  "trx_in_block": 36,
  "virtual_op": 0
}

{
  "block": 22701150,
  "op": [
    "comment",
    {
      "author": "stever82",
      "body": "Hello,\n\nWelcome I am also new to Seemit.  Hoping this site continues to grow.",
      "json_metadata": "{\"tags\":[\"introduceyourself\"],\"app\":\"steemit/0.1\"}",
      "parent_author": "llsourcell",
      "parent_permlink": "introducemyself-welcome-steemit-folks",
      "permlink": "re-llsourcell-introducemyself-welcome-steemit-folks-20180524t042451301z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:24:51",
  "trx_id": "e02cc4c76d16c9eef129dabaa7c0dc00cb7f01e3",
  "trx_in_block": 62,
  "virtual_op": 0
}

{
  "block": 22701138,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "introducemyself-welcome-steemit-folks",
      "voter": "stever82",
      "weight": 10000
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:24:15",
  "trx_id": "d6881f28121f310b604b96aaeef59fc4ea2b7c1d",
  "trx_in_block": 6,
  "virtual_op": 0
}

{
  "block": 22701117,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "introducemyself-welcome-steemit-folks",
      "voter": "sayutimamet",
      "weight": 2500
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:23:12",
  "trx_id": "a0d215cf07f8bbf985fef130f10f39006921e341",
  "trx_in_block": 40,
  "virtual_op": 0
}

{
  "block": 22701114,
  "op": [
    "transfer",
    {
      "amount": "0.002 SBD",
      "from": "earncrypto",
      "memo": "Welcome to Steem @llsourcell, check my blogs for more ways to earn Steem & SBD :)",
      "to": "llsourcell"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:23:03",
  "trx_id": "b60384f7fca98a2e78b50772dc461930813c4cdd",
  "trx_in_block": 62,
  "virtual_op": 0
}

{
  "block": 22701114,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "introducemyself-welcome-steemit-folks",
      "voter": "earncrypto",
      "weight": 4000
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:23:03",
  "trx_id": "0de93ed616a5f954144983db03318b30a5f0f539",
  "trx_in_block": 43,
  "virtual_op": 0
}

{
  "block": 22701093,
  "op": [
    "comment",
    {
      "author": "cryptobryan",
      "body": "Hii, welcome ! familiar with the cryptocurrency world ? here is some free bitcoin mining ! enjoy . if you need help, just send a message to a post https://www.hashmania.net/?ref=213",
      "json_metadata": "{\"tags\":[\"introduceyourself\"],\"links\":[\"https://www.hashmania.net/?ref=213\"],\"app\":\"steemit/0.1\"}",
      "parent_author": "llsourcell",
      "parent_permlink": "introducemyself-welcome-steemit-folks",
      "permlink": "re-llsourcell-introducemyself-welcome-steemit-folks-20180524t042147534z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:22:00",
  "trx_id": "568a05edef8a14405c645c8c87c3ff505b8c2c8f",
  "trx_in_block": 21,
  "virtual_op": 0
}

{
  "block": 22701089,
  "op": [
    "comment",
    {
      "author": "steemplus-bot",
      "body": "#### Welcome to Steem, @llsourcell!\n\nI am a bot coded by the SteemPlus team to help you make the best of your experience on the Steem Blockchain!\nSteemPlus is a Chrome, Opera and Firefox extension that adds tons of features on Steemit.\nIt helps you see the real value of your account, who mentionned you, the value of the votes received, a filtered and sorted feed and much more! All of this in a fast and secure way.\nTo see why **2565 Steemians** use SteemPlus, [install our extension](https://chrome.google.com/webstore/detail/steemplus/mjbkjgcplmaneajhcbegoffkedeankaj?hl=en), read the [documentation](https://github.com/stoodkev/SteemPlus/blob/master/README.md) or the latest release : [SteemPlus on Fundition](/en/@steem-plus/u7pareocg).\n",
      "json_metadata": "{}",
      "parent_author": "llsourcell",
      "parent_permlink": "introducemyself-welcome-steemit-folks",
      "permlink": "introducemyself-welcome-steemit-folks-re-welcome-to-steemplus",
      "title": "Welcome to SteemPlus"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:21:48",
  "trx_id": "8dbd5de649f845f7d9f3fe29f02788582c0f5dfd",
  "trx_in_block": 6,
  "virtual_op": 0
}

{
  "block": 22701066,
  "op": [
    "comment",
    {
      "author": "voluntary-io",
      "body": "Welcome to Steemit @llsourcell :)",
      "json_metadata": "{\"app\":\"insteem/0.1\",\"format\":\"markdown\",\"community\":\"insteem\"}",
      "parent_author": "llsourcell",
      "parent_permlink": "introducemyself-welcome-steemit-folks",
      "permlink": "20180524t042040552z",
      "title": ""
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:20:39",
  "trx_id": "0c5e99b750a0b4acc5a61ad974e13c5fb03859cd",
  "trx_in_block": 70,
  "virtual_op": 0
}

{
  "block": 22701059,
  "op": [
    "vote",
    {
      "author": "llsourcell",
      "permlink": "introducemyself-welcome-steemit-folks",
      "voter": "ax3",
      "weight": 100
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:20:18",
  "trx_id": "850989f68dcdee82e319d8198c7ac4ae73091778",
  "trx_in_block": 62,
  "virtual_op": 0
}

{
  "block": 22701056,
  "op": [
    "comment",
    {
      "author": "llsourcell",
      "body": "![13912677_511886832341680_8630477208311885704_n.jpg](https://cdn.steemitimages.com/DQmXQVYjEXFd1EG4vZ4FbNDmxgZ2JWRbXBsfWW63THRjJxQ/13912677_511886832341680_8630477208311885704_n.jpg)\n\n\n**Hello Steemians**\n\nI am Siraj Raval, Director at School of AI. Youtube Star. Bestselling Author. I'm on a warpath to Inspire and Educate Developers to Build Artificial Intelligence.\n\nMe on social Media: \n\n[Twitter](https://mobile.twitter.com/sirajraval)\n\n[Facebook](https://www.facebook.com/sirajology)\n\n[Youtube](https://www.youtube.com/c/SirajRaval)\n\n[Instragram](https://www.instagram.com/sirajraval/)\n\n\nLove AI, Stay Happy...",
      "json_metadata": "{\"tags\":[\"introduceyourself\",\"introducemyself\",\"steemit\",\"steem\",\"love\"],\"image\":[\"https://cdn.steemitimages.com/DQmXQVYjEXFd1EG4vZ4FbNDmxgZ2JWRbXBsfWW63THRjJxQ/13912677_511886832341680_8630477208311885704_n.jpg\"],\"links\":[\"https://mobile.twitter.com/sirajraval\",\"https://www.facebook.com/sirajology\",\"https://www.youtube.com/c/SirajRaval\",\"https://www.instagram.com/sirajraval/\"],\"app\":\"steemit/0.1\",\"format\":\"markdown\"}",
      "parent_author": "",
      "parent_permlink": "introduceyourself",
      "permlink": "introducemyself-welcome-steemit-folks",
      "title": "Introducemyself: Welcome Steemit Folks"
    }
  ],
  "op_in_trx": 0,
  "timestamp": "2018-05-24T04:20:09",
  "trx_id": "6404fed55547908685fd7291c31e2563f5b91a90",
  "trx_in_block": 37,
  "virtual_op": 0
}