{
  "id": 79656,
  "title": "FYI : Direct Shot Correspondence Matching",
  "url": "/competitions/humpback-whale-identification/discussion/79656",
  "author_name": "hengck23",
  "post_date": "2019-02-06T10:57:55.776000",
  "votes": 5,
  "comment_count": 0,
  "views": 0,
  "content": "<p>An interesting approach:</p>\n\n<p>input  = image + user query region (e.g. attention mask)</p>\n\n<p>output =  retrieved images from a database and found  query regions</p>\n\n<p><a href=\"http://bmvc2018.org/contents/papers/0371.pdf\">http://bmvc2018.org/contents/papers/0371.pdf</a></p>\n\n<p>'Direct Shot Correspondence Matching' - Umer Raf, bmvc2018</p>\n\n<p>How it can be used:</p>\n\n<p>case.1:   say you can detect interesting region in test fluke image (e.g. a cut, a bite marking, etc ...), you can try to find  these regions in the train set. This eliminate false cases of \"new whales\"</p>\n\n<p>case.2: you can hand mark interesting region of selected train images (e.g. these are id without corresponding test images in your current results) . use train+region to query the test and see if you can find the missing whale</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/467023/11191/directshot.png\" alt=\"enter image description here\"></p>",
  "messages": [
    {
      "id": 467023,
      "postDate": "2019-02-06T10:57:55.777Z",
      "content": "<p>An interesting approach:</p>\n\n<p>input  = image + user query region (e.g. attention mask)</p>\n\n<p>output =  retrieved images from a database and found  query regions</p>\n\n<p><a href=\"http://bmvc2018.org/contents/papers/0371.pdf\">http://bmvc2018.org/contents/papers/0371.pdf</a></p>\n\n<p>'Direct Shot Correspondence Matching' - Umer Raf, bmvc2018</p>\n\n<p>How it can be used:</p>\n\n<p>case.1:   say you can detect interesting region in test fluke image (e.g. a cut, a bite marking, etc ...), you can try to find  these regions in the train set. This eliminate false cases of \"new whales\"</p>\n\n<p>case.2: you can hand mark interesting region of selected train images (e.g. these are id without corresponding test images in your current results) . use train+region to query the test and see if you can find the missing whale</p>\n\n<p><img src=\"https://storage.googleapis.com/kaggle-forum-message-attachments/467023/11191/directshot.png\" alt=\"enter image description here\"></p>",
      "rawMarkdown": "An interesting approach:\n\ninput  = image + user query region (e.g. attention mask)\n\noutput =  retrieved images from a database and found  query regions\n\nhttp://bmvc2018.org/contents/papers/0371.pdf\n\n'Direct Shot Correspondence Matching' - Umer Raf, bmvc2018\n\n\nHow it can be used:\n\ncase.1:   say you can detect interesting region in test fluke image (e.g. a cut, a bite marking, etc ...), you can try to find  these regions in the train set. This eliminate false cases of \"new whales\"\n\ncase.2: you can hand mark interesting region of selected train images (e.g. these are id without corresponding test images in your current results) . use train+region to query the test and see if you can find the missing whale\n\n\n   ![enter image description here][1]\n\n\n  [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/467023/11191/directshot.png",
      "votes": 5
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "467023": "An interesting approach:\n\ninput  = image + user query region (e.g. attention mask)\n\noutput =  retrieved images from a database and found  query regions\n\nhttp://bmvc2018.org/contents/papers/0371.pdf\n\n'Direct Shot Correspondence Matching' - Umer Raf, bmvc2018\n\n\nHow it can be used:\n\ncase.1:   say you can detect interesting region in test fluke image (e.g. a cut, a bite marking, etc ...), you can try to find  these regions in the train set. This eliminate false cases of \"new whales\"\n\ncase.2: you can hand mark interesting region of selected train images (e.g. these are id without corresponding test images in your current results) . use train+region to query the test and see if you can find the missing whale\n\n\n   ![enter image description here][1]\n\n\n  [1]: https://storage.googleapis.com/kaggle-forum-message-attachments/467023/11191/directshot.png"
  }
}