{
  "id": 42536,
  "title": "Unclean Data",
  "url": "/competitions/cdiscount-image-classification-challenge/discussion/42536",
  "author_name": "",
  "post_date": "2017-11-01T05:02:33.742521500Z",
  "votes": 4,
  "comment_count": 3,
  "views": 0,
  "content": "<p>After inspection of a few categories it seems that the dataset provided is not very clean. A number of categories contain placeholder images (not a huge problem). Other categories contain images that are mislabeled see 1000000893 - artichokes.</p>\n\n<p>While I realized this is the very problem CDiscount is looking to fix, the overview implied that the training was properly label. Just wanted to make people aware of the issue.</p>",
  "messages": [
    {
      "id": "238304",
      "postDate": "11/01/2017 05:02:33",
      "content": "<p>After inspection of a few categories it seems that the dataset provided is not very clean. A number of categories contain placeholder images (not a huge problem). Other categories contain images that are mislabeled see 1000000893 - artichokes.</p>\n\n<p>While I realized this is the very problem CDiscount is looking to fix, the overview implied that the training was properly label. Just wanted to make people aware of the issue.</p>",
      "rawMarkdown": "After inspection of a few categories it seems that the dataset provided is not very clean. A number of categories contain placeholder images (not a huge problem). Other categories contain images that are mislabeled see 1000000893 - artichokes.\n\nWhile I realized this is the very problem CDiscount is looking to fix, the overview implied that the training was properly label. Just wanted to make people aware of the issue.",
      "votes": null
    },
    {
      "id": "238383",
      "postDate": "11/01/2017 09:53:17",
      "content": "<p>Yeah, I've seen totally mislabelled categories too. For example:</p>\n\n<p>1000012764  AMENAGEMENT URBAIN - VOIRIE AMENAGEMENT URBAIN  ABRI FUMEUR</p>\n\n<p>This category is supposed to contain road signs but it just has pictures of shoes!?</p>\n\n<p>(It's possible I messed up somewhere and put the wrong images into the wrong category when reading the BSON file, but the other categories I looked at mostly seem to be correct.)</p>",
      "rawMarkdown": "Yeah, I've seen totally mislabelled categories too. For example:\n\n1000012764\tAMENAGEMENT URBAIN - VOIRIE\tAMENAGEMENT URBAIN\tABRI FUMEUR\n\nThis category is supposed to contain road signs but it just has pictures of shoes!?\n\n(It's possible I messed up somewhere and put the wrong images into the wrong category when reading the BSON file, but the other categories I looked at mostly seem to be correct.)",
      "votes": null
    },
    {
      "id": "238434",
      "postDate": "11/01/2017 10:34:16",
      "content": "<p>That category seems to contain mostly dress shoes but there are some set top boxes, circuit boards, security camera, running shoes, at least one build a plane and what ever this is. Can we get some guidance from anyone from Kaggle or Discount. Or at least a clarification.</p>",
      "rawMarkdown": "That category seems to contain mostly dress shoes but there are some set top boxes, circuit boards, security camera, running shoes, at least one build a plane and what ever this is. Can we get some guidance from anyone from Kaggle or Discount. Or at least a clarification.",
      "votes": null
    },
    {
      "id": "239426",
      "postDate": "11/03/2017 12:56:02",
      "content": "<p>The dataset sometimes contains mislabeled images in Kaggle. Shortly speaking, they are regarded as noise of data. In StateFarm competition, some images were mislabeled. Please refer below discussion.<br>\n<em>Mislabeled training images</em><br>\n<a href=\"https://www.kaggle.com/c/state-farm-distracted-driver-detection/discussion/2016\">https://www.kaggle.com/c/state-farm-distracted-driver-detection/discussion/2016</a></p>",
      "rawMarkdown": "The dataset sometimes contains mislabeled images in Kaggle. Shortly speaking, they are regarded as noise of data. In StateFarm competition, some images were mislabeled. Please refer below discussion.<br>\n*Mislabeled training images*<br>\nhttps://www.kaggle.com/c/state-farm-distracted-driver-detection/discussion/2016",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 238383,
      "author_name": "humananalog",
      "author_url": "",
      "post_date": "11/01/2017 09:53:17",
      "content": "<p>Yeah, I've seen totally mislabelled categories too. For example:</p>\n\n<p>1000012764  AMENAGEMENT URBAIN - VOIRIE AMENAGEMENT URBAIN  ABRI FUMEUR</p>\n\n<p>This category is supposed to contain road signs but it just has pictures of shoes!?</p>\n\n<p>(It's possible I messed up somewhere and put the wrong images into the wrong category when reading the BSON file, but the other categories I looked at mostly seem to be correct.)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 238434,
      "author_name": "rcodddow",
      "author_url": "",
      "post_date": "11/01/2017 10:34:16",
      "content": "<p>That category seems to contain mostly dress shoes but there are some set top boxes, circuit boards, security camera, running shoes, at least one build a plane and what ever this is. Can we get some guidance from anyone from Kaggle or Discount. Or at least a clarification.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 239426,
      "author_name": "toshik",
      "author_url": "",
      "post_date": "11/03/2017 12:56:02",
      "content": "<p>The dataset sometimes contains mislabeled images in Kaggle. Shortly speaking, they are regarded as noise of data. In StateFarm competition, some images were mislabeled. Please refer below discussion.<br>\n<em>Mislabeled training images</em><br>\n<a href=\"https://www.kaggle.com/c/state-farm-distracted-driver-detection/discussion/2016\">https://www.kaggle.com/c/state-farm-distracted-driver-detection/discussion/2016</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "238304": "After inspection of a few categories it seems that the dataset provided is not very clean. A number of categories contain placeholder images (not a huge problem). Other categories contain images that are mislabeled see 1000000893 - artichokes.\n\nWhile I realized this is the very problem CDiscount is looking to fix, the overview implied that the training was properly label. Just wanted to make people aware of the issue.",
    "238383": "Yeah, I've seen totally mislabelled categories too. For example:\n\n1000012764\tAMENAGEMENT URBAIN - VOIRIE\tAMENAGEMENT URBAIN\tABRI FUMEUR\n\nThis category is supposed to contain road signs but it just has pictures of shoes!?\n\n(It's possible I messed up somewhere and put the wrong images into the wrong category when reading the BSON file, but the other categories I looked at mostly seem to be correct.)",
    "238434": "That category seems to contain mostly dress shoes but there are some set top boxes, circuit boards, security camera, running shoes, at least one build a plane and what ever this is. Can we get some guidance from anyone from Kaggle or Discount. Or at least a clarification.",
    "239426": "The dataset sometimes contains mislabeled images in Kaggle. Shortly speaking, they are regarded as noise of data. In StateFarm competition, some images were mislabeled. Please refer below discussion.<br>\n*Mislabeled training images*<br>\nhttps://www.kaggle.com/c/state-farm-distracted-driver-detection/discussion/2016"
  },
  "source": "meta"
}