{
  "id": 26542,
  "title": "Possible Deep Learning Starting Point?",
  "url": "/competitions/dstl-satellite-imagery-feature-detection/discussion/26542",
  "author_name": "broken_alchemy",
  "post_date": "2016-12-15T19:16:18.283000",
  "votes": 19,
  "comment_count": 7,
  "views": 1847,
  "content": "<p>Hey,</p>\n\n<p>I recently read <a href=\"https://tryolabs.com/blog/2016/12/06/major-advancements-deep-learning-2016/\" title=\"The major advancements in Deep Learning in 2016\">The major advancements in Deep Learning in 2016</a>. The advancement that seems applicable is <a href=\"https://phillipi.github.io/pix2pix/\" title=\"Image-to-Image Translation with Conditional Adversarial Nets\">Image-to-Image Translation with Conditional Adversarial Nets</a> (link contains paper and code solution using torch)</p>\n\n<p>Example Results (Note labels to street scene):\n<img src=\"https://phillipi.github.io/pix2pix/images/teaser_v3.jpg\" alt=\"Example Results\" title=\"Example Results\"></p>\n\n<p>This competition seems like an exciting reason to learn more about Generative Adversarial Networks (GANs)!</p>",
  "messages": [
    {
      "id": 150557,
      "postDate": "2016-12-15T19:16:18.283Z",
      "content": "<p>Hey,</p>\n\n<p>I recently read <a href=\"https://tryolabs.com/blog/2016/12/06/major-advancements-deep-learning-2016/\" title=\"The major advancements in Deep Learning in 2016\">The major advancements in Deep Learning in 2016</a>. The advancement that seems applicable is <a href=\"https://phillipi.github.io/pix2pix/\" title=\"Image-to-Image Translation with Conditional Adversarial Nets\">Image-to-Image Translation with Conditional Adversarial Nets</a> (link contains paper and code solution using torch)</p>\n\n<p>Example Results (Note labels to street scene):\n<img src=\"https://phillipi.github.io/pix2pix/images/teaser_v3.jpg\" alt=\"Example Results\" title=\"Example Results\"></p>\n\n<p>This competition seems like an exciting reason to learn more about Generative Adversarial Networks (GANs)!</p>",
      "rawMarkdown": "Hey,\r\n\r\nI recently read [The major advancements in Deep Learning in 2016][1]. The advancement that seems applicable is [Image-to-Image Translation with Conditional Adversarial Nets][2] (link contains paper and code solution using torch)\r\n\r\nExample Results (Note labels to street scene):\r\n![Example Results][3]\r\n\r\nThis competition seems like an exciting reason to learn more about Generative Adversarial Networks (GANs)!\r\n\r\n\r\n  [1]: https://tryolabs.com/blog/2016/12/06/major-advancements-deep-learning-2016/ \"The major advancements in Deep Learning in 2016\"\r\n  [2]: https://phillipi.github.io/pix2pix/ \"Image-to-Image Translation with Conditional Adversarial Nets\"\r\n  [3]: https://phillipi.github.io/pix2pix/images/teaser_v3.jpg \"Example Results\"",
      "votes": 19
    },
    {
      "id": 151318,
      "postDate": "2016-12-20T03:15:13.323Z",
      "content": "<p>Sounds good at first sight.</p>\n\n<p>Image generation with GAN seems real. But there are two points that may cause GAN not very suitable for the dstl task:\n   1, dstl is multi-class labeling, one pixel may be included in multiple polygons;\n   2, the output is precise polygon vertices coordinates. Labeling output image of GAN needs further grouping.</p>\n\n<p>Still worth a shot!</p>",
      "rawMarkdown": "Sounds good at first sight.\r\n\r\nImage generation with GAN seems real. But there are two points that may cause GAN not very suitable for the dstl task:\r\n   1, dstl is multi-class labeling, one pixel may be included in multiple polygons;\r\n   2, the output is precise polygon vertices coordinates. Labeling output image of GAN needs further grouping.\r\n\r\nStill worth a shot!"
    },
    {
      "id": 150632,
      "postDate": "2016-12-16T04:49:25.457Z",
      "content": "<p>Super cool, can't fill in my schedule though:(, will try!</p>",
      "rawMarkdown": "Super cool, can't fill in my schedule though:(, will try!"
    },
    {
      "id": 150847,
      "postDate": "2016-12-17T02:00:11.513Z",
      "rawMarkdown": "",
      "isDeleted": true
    },
    {
      "id": 151457,
      "postDate": "2016-12-20T21:59:05.117Z",
      "content": "<p>informative read, Thanks !!!</p>",
      "rawMarkdown": "informative read, Thanks !!!"
    },
    {
      "id": 150828,
      "postDate": "2016-12-16T22:32:37.073Z",
      "content": "<p>Very Interesting! Thanks!</p>",
      "rawMarkdown": "Very Interesting! Thanks!"
    },
    {
      "id": 150620,
      "postDate": "2016-12-16T02:25:50.227Z",
      "content": "<p>powerful, thanks for share!</p>",
      "rawMarkdown": "powerful, thanks for share!"
    },
    {
      "id": 150559,
      "postDate": "2016-12-15T19:29:58.257Z",
      "content": "<p>Very nice article, thanks!</p>",
      "rawMarkdown": "Very nice article, thanks!"
    }
  ],
  "comments": [
    {
      "id": 151318,
      "author_name": "JunlongLiu",
      "author_url": "",
      "post_date": "2016-12-20T03:15:13.323000",
      "content": "<p>Sounds good at first sight.</p>\n\n<p>Image generation with GAN seems real. But there are two points that may cause GAN not very suitable for the dstl task:\n   1, dstl is multi-class labeling, one pixel may be included in multiple polygons;\n   2, the output is precise polygon vertices coordinates. Labeling output image of GAN needs further grouping.</p>\n\n<p>Still worth a shot!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 150632,
      "author_name": "Hang(Yohan) Yu",
      "author_url": "",
      "post_date": "2016-12-16T04:49:25.457000",
      "content": "<p>Super cool, can't fill in my schedule though:(, will try!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 150847,
      "author_name": "",
      "author_url": "",
      "post_date": "2016-12-17T02:00:11.513000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 151457,
      "author_name": "Deependra Mishra",
      "author_url": "",
      "post_date": "2016-12-20T21:59:05.117000",
      "content": "<p>informative read, Thanks !!!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 150828,
      "author_name": "LEE-SL",
      "author_url": "",
      "post_date": "2016-12-16T22:32:37.073000",
      "content": "<p>Very Interesting! Thanks!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 150620,
      "author_name": "kbulls",
      "author_url": "",
      "post_date": "2016-12-16T02:25:50.227000",
      "content": "<p>powerful, thanks for share!</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 150559,
      "author_name": "MTDzi",
      "author_url": "",
      "post_date": "2016-12-15T19:29:58.257000",
      "content": "<p>Very nice article, thanks!</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "150557": "Hey,\r\n\r\nI recently read [The major advancements in Deep Learning in 2016][1]. The advancement that seems applicable is [Image-to-Image Translation with Conditional Adversarial Nets][2] (link contains paper and code solution using torch)\r\n\r\nExample Results (Note labels to street scene):\r\n![Example Results][3]\r\n\r\nThis competition seems like an exciting reason to learn more about Generative Adversarial Networks (GANs)!\r\n\r\n\r\n  [1]: https://tryolabs.com/blog/2016/12/06/major-advancements-deep-learning-2016/ \"The major advancements in Deep Learning in 2016\"\r\n  [2]: https://phillipi.github.io/pix2pix/ \"Image-to-Image Translation with Conditional Adversarial Nets\"\r\n  [3]: https://phillipi.github.io/pix2pix/images/teaser_v3.jpg \"Example Results\"",
    "151318": "Sounds good at first sight.\r\n\r\nImage generation with GAN seems real. But there are two points that may cause GAN not very suitable for the dstl task:\r\n   1, dstl is multi-class labeling, one pixel may be included in multiple polygons;\r\n   2, the output is precise polygon vertices coordinates. Labeling output image of GAN needs further grouping.\r\n\r\nStill worth a shot!",
    "150632": "Super cool, can't fill in my schedule though:(, will try!",
    "150847": "",
    "151457": "informative read, Thanks !!!",
    "150828": "Very Interesting! Thanks!",
    "150620": "powerful, thanks for share!",
    "150559": "Very nice article, thanks!"
  }
}