{
  "id": 94726,
  "title": "Has anyone analysed how many images are 'poorly' labeled?",
  "url": "/competitions/imaterialist-fashion-2019-FGVC6/discussion/94726",
  "author_name": "cherring",
  "post_date": "2019-06-06T13:27:53.688000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I use the term 'poorly' very loosely.</p>\n\n<p>I know that part of the challenge is to try and learn what you should detect and what you shouldn't. That is all part of the fun :)</p>\n\n<p>But I compare some of my predictions to ground truths and I am not sure I can expect too much more from it.\nTake this image for example:\n&gt; 000cd2e13d1bdd28f480304d7bb9e1ca</p>\n\n<p>This is the gt:\n<a href=\"https://imgur.com/a/Ou8fcWJ\">https://imgur.com/a/Ou8fcWJ</a></p>\n\n<p>This is my prediction:\n<a href=\"https://imgur.com/a/MWd3qpn\">https://imgur.com/a/MWd3qpn</a></p>\n\n<p>I have my little model a pat on the back and a tasty biscuit for that :) I am sure if I have lots more time and resources to work on this then I could do something smarter and somehow teach it to only detect the girl on the left.</p>\n\n<p>But has anyone done any analysis on just how many images have these false positives?\nWell I guess they are more false true positives....</p>\n\n<p>I can't wait for a couple days when the competition is over - will be very interesting to see what smart stuff everyone else has done :)</p>",
  "messages": [
    {
      "id": 546329,
      "postDate": "2019-06-06T13:27:53.687Z",
      "content": "<p>I use the term 'poorly' very loosely.</p>\n\n<p>I know that part of the challenge is to try and learn what you should detect and what you shouldn't. That is all part of the fun :)</p>\n\n<p>But I compare some of my predictions to ground truths and I am not sure I can expect too much more from it.\nTake this image for example:\n&gt; 000cd2e13d1bdd28f480304d7bb9e1ca</p>\n\n<p>This is the gt:\n<a href=\"https://imgur.com/a/Ou8fcWJ\">https://imgur.com/a/Ou8fcWJ</a></p>\n\n<p>This is my prediction:\n<a href=\"https://imgur.com/a/MWd3qpn\">https://imgur.com/a/MWd3qpn</a></p>\n\n<p>I have my little model a pat on the back and a tasty biscuit for that :) I am sure if I have lots more time and resources to work on this then I could do something smarter and somehow teach it to only detect the girl on the left.</p>\n\n<p>But has anyone done any analysis on just how many images have these false positives?\nWell I guess they are more false true positives....</p>\n\n<p>I can't wait for a couple days when the competition is over - will be very interesting to see what smart stuff everyone else has done :)</p>",
      "rawMarkdown": "I use the term 'poorly' very loosely.\n\nI know that part of the challenge is to try and learn what you should detect and what you shouldn't. That is all part of the fun :)\n\nBut I compare some of my predictions to ground truths and I am not sure I can expect too much more from it.\nTake this image for example:\n&gt; 000cd2e13d1bdd28f480304d7bb9e1ca\n\nThis is the gt:\nhttps://imgur.com/a/Ou8fcWJ\n\nThis is my prediction:\nhttps://imgur.com/a/MWd3qpn\n\nI have my little model a pat on the back and a tasty biscuit for that :) I am sure if I have lots more time and resources to work on this then I could do something smarter and somehow teach it to only detect the girl on the left.\n\nBut has anyone done any analysis on just how many images have these false positives?\nWell I guess they are more false true positives....\n\nI can't wait for a couple days when the competition is over - will be very interesting to see what smart stuff everyone else has done :)\n\n",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "546329": "I use the term 'poorly' very loosely.\n\nI know that part of the challenge is to try and learn what you should detect and what you shouldn't. That is all part of the fun :)\n\nBut I compare some of my predictions to ground truths and I am not sure I can expect too much more from it.\nTake this image for example:\n&gt; 000cd2e13d1bdd28f480304d7bb9e1ca\n\nThis is the gt:\nhttps://imgur.com/a/Ou8fcWJ\n\nThis is my prediction:\nhttps://imgur.com/a/MWd3qpn\n\nI have my little model a pat on the back and a tasty biscuit for that :) I am sure if I have lots more time and resources to work on this then I could do something smarter and somehow teach it to only detect the girl on the left.\n\nBut has anyone done any analysis on just how many images have these false positives?\nWell I guess they are more false true positives....\n\nI can't wait for a couple days when the competition is over - will be very interesting to see what smart stuff everyone else has done :)\n\n"
  }
}