{
  "id": 38926,
  "title": "predict labels in groups of nxn pixels instead of single pixel",
  "url": "/competitions/carvana-image-masking-challenge/discussion/38926",
  "author_name": "",
  "post_date": "2017-09-03T14:56:15.304848200Z",
  "votes": 5,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Here is an idea to share. I haven't try it it yet and not sure if it will work. It capture co-occurrences and output a lower resolution feature map. If you want, you can have overlap blocks and ensemble the results at a pixel location</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/218303/7237/encode.png\" alt=\"enter image description here\" title=\"\"></p>",
  "messages": [
    {
      "id": "218303",
      "postDate": "09/03/2017 14:56:15",
      "content": "<p>Here is an idea to share. I haven't try it it yet and not sure if it will work. It capture co-occurrences and output a lower resolution feature map. If you want, you can have overlap blocks and ensemble the results at a pixel location</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/218303/7237/encode.png\" alt=\"enter image description here\" title=\"\"></p>",
      "rawMarkdown": "Here is an idea to share. I haven't try it it yet and not sure if it will work. It capture co-occurrences and output a lower resolution feature map. If you want, you can have overlap blocks and ensemble the results at a pixel location\n\n  ![enter image description here][1]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/218303/7237/encode.png",
      "votes": null
    },
    {
      "id": "218336",
      "postDate": "09/03/2017 19:10:46",
      "content": "<p>Interesting. One idea related to focusing on the borders: <a href=\"http://www.cs.toronto.edu/polyrnn/\">http://www.cs.toronto.edu/polyrnn/</a></p>",
      "rawMarkdown": "Interesting. One idea related to focusing on the borders: http://www.cs.toronto.edu/polyrnn/",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 218336,
      "author_name": "antorsae",
      "author_url": "",
      "post_date": "09/03/2017 19:10:46",
      "content": "<p>Interesting. One idea related to focusing on the borders: <a href=\"http://www.cs.toronto.edu/polyrnn/\">http://www.cs.toronto.edu/polyrnn/</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "218303": "Here is an idea to share. I haven't try it it yet and not sure if it will work. It capture co-occurrences and output a lower resolution feature map. If you want, you can have overlap blocks and ensemble the results at a pixel location\n\n  ![enter image description here][1]\n\n\n  [1]: https://kaggle2.blob.core.windows.net/forum-message-attachments/218303/7237/encode.png",
    "218336": "Interesting. One idea related to focusing on the borders: http://www.cs.toronto.edu/polyrnn/"
  },
  "source": "meta"
}