{
  "id": 456663,
  "title": "Pictures without eyes",
  "url": "/competitions/applications-of-deep-learning-wustl-fall-2023/discussion/456663",
  "author_name": "",
  "post_date": "2023-11-21T06:35:38.310749100Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I noticed, that there are some pictures in the training dataset without glasses on it, but still counting as pictures with glasses. I just wanted to ask how you handle this. I was thinking of kicking them out of the dataset to not create confusion whilst training, but I am quite sure, that there are pictures of the same kind in the Test dataset.</p>\n<p>Here is what I mean:<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17168218%2F2651e030fe3681b8ea6a557ec3eb8cbb%2Fkaggle_1.png?generation=1700547872116963&amp;alt=media\" alt=\"\"></p>\n<p>I trained this model only for one epoch just so I can show you what I mean. Also there are many more cases like this where the persons eyes are not visible. </p>",
  "messages": [
    {
      "id": "2532559",
      "postDate": "11/21/2023 06:35:38",
      "content": "<p>I noticed, that there are some pictures in the training dataset without glasses on it, but still counting as pictures with glasses. I just wanted to ask how you handle this. I was thinking of kicking them out of the dataset to not create confusion whilst training, but I am quite sure, that there are pictures of the same kind in the Test dataset.</p>\n<p>Here is what I mean:<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17168218%2F2651e030fe3681b8ea6a557ec3eb8cbb%2Fkaggle_1.png?generation=1700547872116963&amp;alt=media\" alt=\"\"></p>\n<p>I trained this model only for one epoch just so I can show you what I mean. Also there are many more cases like this where the persons eyes are not visible. </p>",
      "rawMarkdown": "I noticed, that there are some pictures in the training dataset without glasses on it, but still counting as pictures with glasses. I just wanted to ask how you handle this. I was thinking of kicking them out of the dataset to not create confusion whilst training, but I am quite sure, that there are pictures of the same kind in the Test dataset.\n\nHere is what I mean:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17168218%2F2651e030fe3681b8ea6a557ec3eb8cbb%2Fkaggle_1.png?generation=1700547872116963&alt=media)\n\nI trained this model only for one epoch just so I can show you what I mean. Also there are many more cases like this where the persons eyes are not visible.",
      "votes": null
    },
    {
      "id": "2539272",
      "postDate": "11/26/2023 23:27:44",
      "content": "<p>Pros and cons of each. If you them in, there will be such photos in the test set, and maybe there are  clues in the images to predict glasses even if you can't see eyes. If you take them out, maybe there is no additional information and it will give you a better train.</p>",
      "rawMarkdown": "Pros and cons of each. If you them in, there will be such photos in the test set, and maybe there are  clues in the images to predict glasses even if you can't see eyes. If you take them out, maybe there is no additional information and it will give you a better train.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2539272,
      "author_name": "jeffheaton",
      "author_url": "",
      "post_date": "11/26/2023 23:27:44",
      "content": "<p>Pros and cons of each. If you them in, there will be such photos in the test set, and maybe there are  clues in the images to predict glasses even if you can't see eyes. If you take them out, maybe there is no additional information and it will give you a better train.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2532559": "I noticed, that there are some pictures in the training dataset without glasses on it, but still counting as pictures with glasses. I just wanted to ask how you handle this. I was thinking of kicking them out of the dataset to not create confusion whilst training, but I am quite sure, that there are pictures of the same kind in the Test dataset.\n\nHere is what I mean:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F17168218%2F2651e030fe3681b8ea6a557ec3eb8cbb%2Fkaggle_1.png?generation=1700547872116963&alt=media)\n\nI trained this model only for one epoch just so I can show you what I mean. Also there are many more cases like this where the persons eyes are not visible.",
    "2539272": "Pros and cons of each. If you them in, there will be such photos in the test set, and maybe there are  clues in the images to predict glasses even if you can't see eyes. If you take them out, maybe there is no additional information and it will give you a better train."
  },
  "source": "meta"
}