{
  "id": 77622,
  "title": "Whale GAN, generating fake whales",
  "url": "/competitions/humpback-whale-identification/discussion/77622",
  "author_name": "Brian",
  "post_date": "2019-01-15T02:23:08.932000",
  "votes": 12,
  "comment_count": 2,
  "views": 0,
  "content": "<p>One thing I am exploring is a GAN to augment the whale data. I've posted the kernel up here: <a href=\"https://www.kaggle.com/ldm314/whale-gan\">Whale GAN</a></p>\n\n<p>It takes some time to run on a kernel, but locally it is making some pretty interesting images at 30000+ iterations. Attached is the result from 39000 iterations. \n<img src=\"https://i.imgur.com/kBhjc7o.jpg\" alt=\"Whale Gan Image\"></p>",
  "messages": [
    {
      "id": 456033,
      "postDate": "2019-01-15T02:23:08.933Z",
      "content": "<p>One thing I am exploring is a GAN to augment the whale data. I've posted the kernel up here: <a href=\"https://www.kaggle.com/ldm314/whale-gan\">Whale GAN</a></p>\n\n<p>It takes some time to run on a kernel, but locally it is making some pretty interesting images at 30000+ iterations. Attached is the result from 39000 iterations. \n<img src=\"https://i.imgur.com/kBhjc7o.jpg\" alt=\"Whale Gan Image\"></p>",
      "rawMarkdown": "One thing I am exploring is a GAN to augment the whale data. I've posted the kernel up here: [Whale GAN][1]\n\nIt takes some time to run on a kernel, but locally it is making some pretty interesting images at 30000+ iterations. Attached is the result from 39000 iterations. \n![Whale Gan Image][2]\n\n\n  [1]: https://www.kaggle.com/ldm314/whale-gan\n  [2]: https://i.imgur.com/kBhjc7o.jpg",
      "votes": 12
    },
    {
      "id": 456090,
      "postDate": "2019-01-15T05:49:33.397Z",
      "rawMarkdown": "",
      "isDeleted": true,
      "replies": [
        {
          "id": 456096,
          "postDate": "2019-01-15T06:13:30.127Z",
          "content": "<p>These whales don't match any particular ID. The trained network takes an input of random noise shaped ( 4250 ) and outputs \"whale like\" images at 384x384. I will then use that to train a network that takes a whale input and outputs the ( 4250 ) sized embedding.</p>\n\n<p>Will this do anything useful? So far it has been useful for myself to learn about GAN networks. I hope it will also be useful in this contest somehow but I am not sure on that.</p>",
          "rawMarkdown": "These whales don't match any particular ID. The trained network takes an input of random noise shaped ( 4250 ) and outputs \"whale like\" images at 384x384. I will then use that to train a network that takes a whale input and outputs the ( 4250 ) sized embedding.\n\nWill this do anything useful? So far it has been useful for myself to learn about GAN networks. I hope it will also be useful in this contest somehow but I am not sure on that.",
          "votes": 3
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 456090,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-01-15T05:49:33.397000",
      "content": "",
      "votes": 0,
      "replies": [
        {
          "id": 456096,
          "author_name": "Brian",
          "author_url": "",
          "post_date": "2019-01-15T06:13:30.127000",
          "content": "<p>These whales don't match any particular ID. The trained network takes an input of random noise shaped ( 4250 ) and outputs \"whale like\" images at 384x384. I will then use that to train a network that takes a whale input and outputs the ( 4250 ) sized embedding.</p>\n\n<p>Will this do anything useful? So far it has been useful for myself to learn about GAN networks. I hope it will also be useful in this contest somehow but I am not sure on that.</p>",
          "votes": 3,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "456033": "One thing I am exploring is a GAN to augment the whale data. I've posted the kernel up here: [Whale GAN][1]\n\nIt takes some time to run on a kernel, but locally it is making some pretty interesting images at 30000+ iterations. Attached is the result from 39000 iterations. \n![Whale Gan Image][2]\n\n\n  [1]: https://www.kaggle.com/ldm314/whale-gan\n  [2]: https://i.imgur.com/kBhjc7o.jpg",
    "456090": ""
  }
}