{
  "id": 22297,
  "title": "Noisy labels in the train set",
  "url": "/competitions/state-farm-distracted-driver-detection/discussion/22297",
  "author_name": "",
  "post_date": "2016-07-15T13:15:10.480Z",
  "votes": 1,
  "comment_count": 2,
  "views": 446,
  "content": "<p>As someone pointed out in another topic, class 0 (safe driving) gets confused with class 9 (talking with passengers) quite often. Taking a close look at these classes, naturally they are fairly overlapped with many confused cases. For example, the image of driver p081 below is classified as safe driving, which doesn't look like the case to me. </p>\n\n<p><img src=\"https://www.kaggle.com/blobs/download/forum-message-attachment-files/4605/p081_img_87245.jpg\" alt=\"enter image description here\" title></p>\n\n<p>Whereas the driver below is classified as class c9, which also doesn't seem like the case.</p>\n\n<p><img src=\"https://www.kaggle.com/blobs/download/forum-message-attachment-files/4606/p002_img_18997.jpg\" alt=\"enter image description here\" title></p>\n\n<p>This image is classified as C9, but it should probably belong to C8 (hair makeup)</p>\n\n<p><img src=\"https://www.kaggle.com/blobs/download/forum-message-attachment-files/4607/p081_img_34622.jpg\" alt=\"enter image description here\" title></p>\n\n<p>I can spot quite a number of other examples.</p>\n\n<p>This leads to the question, whether we should take some effort in cleaning up the training set.</p>",
  "messages": [
    {
      "id": "127807",
      "postDate": "07/15/2016 13:15:10",
      "content": "<p>As someone pointed out in another topic, class 0 (safe driving) gets confused with class 9 (talking with passengers) quite often. Taking a close look at these classes, naturally they are fairly overlapped with many confused cases. For example, the image of driver p081 below is classified as safe driving, which doesn't look like the case to me. </p>\n\n<p><img src=\"https://www.kaggle.com/blobs/download/forum-message-attachment-files/4605/p081_img_87245.jpg\" alt=\"enter image description here\" title></p>\n\n<p>Whereas the driver below is classified as class c9, which also doesn't seem like the case.</p>\n\n<p><img src=\"https://www.kaggle.com/blobs/download/forum-message-attachment-files/4606/p002_img_18997.jpg\" alt=\"enter image description here\" title></p>\n\n<p>This image is classified as C9, but it should probably belong to C8 (hair makeup)</p>\n\n<p><img src=\"https://www.kaggle.com/blobs/download/forum-message-attachment-files/4607/p081_img_34622.jpg\" alt=\"enter image description here\" title></p>\n\n<p>I can spot quite a number of other examples.</p>\n\n<p>This leads to the question, whether we should take some effort in cleaning up the training set.</p>",
      "rawMarkdown": "As someone pointed out in another topic, class 0 (safe driving) gets confused with class 9 (talking with passengers) quite often. Taking a close look at these classes, naturally they are fairly overlapped with many confused cases. For example, the image of driver p081 below is classified as safe driving, which doesn't look like the case to me. \r\n\r\n![enter image description here][1]\r\n\r\nWhereas the driver below is classified as class c9, which also doesn't seem like the case.\r\n\r\n![enter image description here][2]\r\n\r\nThis image is classified as C9, but it should probably belong to C8 (hair makeup)\r\n\r\n![enter image description here][3]\r\n\r\nI can spot quite a number of other examples.\r\n\r\nThis leads to the question, whether we should take some effort in cleaning up the training set.\r\n\r\n\r\n  [1]: https://www.kaggle.com/blobs/download/forum-message-attachment-files/4605/p081_img_87245.jpg\r\n  [2]: https://www.kaggle.com/blobs/download/forum-message-attachment-files/4606/p002_img_18997.jpg\r\n  [3]: https://www.kaggle.com/blobs/download/forum-message-attachment-files/4607/p081_img_34622.jpg",
      "votes": null
    },
    {
      "id": "127818",
      "postDate": "07/15/2016 14:28:34",
      "content": "<p>This was already pointed out <a href=\"https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20162/mislabeled-training-images\">https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20162/mislabeled-training-images</a>.\nPersonally, I wouldn't recommend it, as both training and testing data were collected in a similar fashion, they will probably both be similarly noisy.</p>",
      "rawMarkdown": "This was already pointed out https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20162/mislabeled-training-images.\r\nPersonally, I wouldn't recommend it, as both training and testing data were collected in a similar fashion, they will probably both be similarly noisy.",
      "votes": null
    },
    {
      "id": "127846",
      "postDate": "07/15/2016 17:33:05",
      "content": "<p>IMO ALL of the first lady's images should be removed from the training set. I don't think she followed instructions at all. She must have been smitten by the passenger because she was gazing into the passengers eye in all her images. </p>",
      "rawMarkdown": "IMO ALL of the first lady's images should be removed from the training set. I don't think she followed instructions at all. She must have been smitten by the passenger because she was gazing into the passengers eye in all her images.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 127818,
      "author_name": "bobutis",
      "author_url": "",
      "post_date": "07/15/2016 14:28:34",
      "content": "<p>This was already pointed out <a href=\"https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20162/mislabeled-training-images\">https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20162/mislabeled-training-images</a>.\nPersonally, I wouldn't recommend it, as both training and testing data were collected in a similar fashion, they will probably both be similarly noisy.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 127846,
      "author_name": "godaibo",
      "author_url": "",
      "post_date": "07/15/2016 17:33:05",
      "content": "<p>IMO ALL of the first lady's images should be removed from the training set. I don't think she followed instructions at all. She must have been smitten by the passenger because she was gazing into the passengers eye in all her images. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "127807": "As someone pointed out in another topic, class 0 (safe driving) gets confused with class 9 (talking with passengers) quite often. Taking a close look at these classes, naturally they are fairly overlapped with many confused cases. For example, the image of driver p081 below is classified as safe driving, which doesn't look like the case to me. \r\n\r\n![enter image description here][1]\r\n\r\nWhereas the driver below is classified as class c9, which also doesn't seem like the case.\r\n\r\n![enter image description here][2]\r\n\r\nThis image is classified as C9, but it should probably belong to C8 (hair makeup)\r\n\r\n![enter image description here][3]\r\n\r\nI can spot quite a number of other examples.\r\n\r\nThis leads to the question, whether we should take some effort in cleaning up the training set.\r\n\r\n\r\n  [1]: https://www.kaggle.com/blobs/download/forum-message-attachment-files/4605/p081_img_87245.jpg\r\n  [2]: https://www.kaggle.com/blobs/download/forum-message-attachment-files/4606/p002_img_18997.jpg\r\n  [3]: https://www.kaggle.com/blobs/download/forum-message-attachment-files/4607/p081_img_34622.jpg",
    "127818": "This was already pointed out https://www.kaggle.com/c/state-farm-distracted-driver-detection/forums/t/20162/mislabeled-training-images.\r\nPersonally, I wouldn't recommend it, as both training and testing data were collected in a similar fashion, they will probably both be similarly noisy.",
    "127846": "IMO ALL of the first lady's images should be removed from the training set. I don't think she followed instructions at all. She must have been smitten by the passenger because she was gazing into the passengers eye in all her images."
  },
  "source": "meta"
}