{
  "id": 224294,
  "title": "Is manual annotation of training set allowed?",
  "url": "/competitions/bms-molecular-translation/discussion/224294",
  "author_name": "Thomas SELECK",
  "post_date": "2021-03-07T17:57:09.713000",
  "votes": 4,
  "comment_count": 8,
  "views": 0,
  "content": "<p>Hello everyone,</p>\n<p>Can you tell me if doing manual annotation of training set is allowed? Doing this can be useful to train object detectors, for instance.</p>",
  "messages": [
    {
      "id": 1229966,
      "postDate": "2021-03-07T18:09:18.097Z",
      "content": "<p>I would presume it is allowed, as it has been in previous Kaggle competitions. But note, it IS against the rules to hand label anything from the test set.</p>",
      "rawMarkdown": "I would presume it is allowed, as it has been in previous Kaggle competitions. But note, it IS against the rules to hand label anything from the test set.",
      "votes": 4,
      "replies": [
        {
          "id": 1229971,
          "postDate": "2021-03-07T18:14:55.443Z",
          "rawMarkdown": "",
          "isDeleted": true
        },
        {
          "id": 1230095,
          "postDate": "2021-03-07T19:16:40.213Z",
          "content": "<p>Can I do absolutely anything with the test set beside hand labeling them (a biological being looking at it and creating a label for it)?<br>\nSo can I for example create soft labels for it or using them in a self supervised pretraining or something like these?</p>",
          "rawMarkdown": "Can I do absolutely anything with the test set beside hand labeling them (a biological being looking at it and creating a label for it)?\nSo can I for example create soft labels for it or using them in a self supervised pretraining or something like these?"
        },
        {
          "id": 1230109,
          "postDate": "2021-03-07T19:31:12.990Z",
          "content": "<p>Like pseudo-labeling? Yes, that should be allowed.</p>",
          "rawMarkdown": "Like pseudo-labeling? Yes, that should be allowed.",
          "votes": 2
        },
        {
          "id": 1230121,
          "postDate": "2021-03-07T19:46:39.403Z",
          "content": "<p>Thank you! (yes, like pseudo-labeling and like SimCLR and the rest)</p>",
          "rawMarkdown": "Thank you! (yes, like pseudo-labeling and like SimCLR and the rest)"
        }
      ]
    },
    {
      "id": 1229953,
      "postDate": "2021-03-07T17:57:09.713Z",
      "content": "<p>Hello everyone,</p>\n<p>Can you tell me if doing manual annotation of training set is allowed? Doing this can be useful to train object detectors, for instance.</p>",
      "rawMarkdown": "Hello everyone,\n\nCan you tell me if doing manual annotation of training set is allowed? Doing this can be useful to train object detectors, for instance.",
      "votes": 4
    },
    {
      "id": 1244704,
      "postDate": "2021-03-19T06:58:44.620Z",
      "content": "<p>Hi, did you end up annotating the training data? If so, how do you go about doing it? Are there any automated way of doing it?</p>",
      "rawMarkdown": "Hi, did you end up annotating the training data? If so, how do you go about doing it? Are there any automated way of doing it?",
      "replies": [
        {
          "id": 1247067,
          "postDate": "2021-03-21T11:24:33.627Z",
          "content": "<p>My idea is to extract a set of connected components from each image automatically from the training set and then classify them using a simple classification algorithm to extract letters. For the moment, I just annotated about a thousand letters.</p>",
          "rawMarkdown": "My idea is to extract a set of connected components from each image automatically from the training set and then classify them using a simple classification algorithm to extract letters. For the moment, I just annotated about a thousand letters.",
          "votes": 1
        }
      ]
    },
    {
      "id": 1230717,
      "postDate": "2021-03-08T11:42:29.260Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1229966,
      "author_name": "Matt",
      "author_url": "",
      "post_date": "2021-03-07T18:09:18.097000",
      "content": "<p>I would presume it is allowed, as it has been in previous Kaggle competitions. But note, it IS against the rules to hand label anything from the test set.</p>",
      "votes": 4,
      "replies": [
        {
          "id": 1229971,
          "author_name": "",
          "author_url": "",
          "post_date": "2021-03-07T18:14:55.443000",
          "content": "",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1230095,
          "author_name": "nofreewill42",
          "author_url": "",
          "post_date": "2021-03-07T19:16:40.213000",
          "content": "<p>Can I do absolutely anything with the test set beside hand labeling them (a biological being looking at it and creating a label for it)?<br>\nSo can I for example create soft labels for it or using them in a self supervised pretraining or something like these?</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1230109,
          "author_name": "Matt",
          "author_url": "",
          "post_date": "2021-03-07T19:31:12.990000",
          "content": "<p>Like pseudo-labeling? Yes, that should be allowed.</p>",
          "votes": 2,
          "replies": []
        },
        {
          "id": 1230121,
          "author_name": "nofreewill42",
          "author_url": "",
          "post_date": "2021-03-07T19:46:39.403000",
          "content": "<p>Thank you! (yes, like pseudo-labeling and like SimCLR and the rest)</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1244704,
      "author_name": "JunYong Tong",
      "author_url": "",
      "post_date": "2021-03-19T06:58:44.620000",
      "content": "<p>Hi, did you end up annotating the training data? If so, how do you go about doing it? Are there any automated way of doing it?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 1247067,
          "author_name": "Thomas SELECK",
          "author_url": "",
          "post_date": "2021-03-21T11:24:33.627000",
          "content": "<p>My idea is to extract a set of connected components from each image automatically from the training set and then classify them using a simple classification algorithm to extract letters. For the moment, I just annotated about a thousand letters.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 1230717,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-03-08T11:42:29.260000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1229966": "I would presume it is allowed, as it has been in previous Kaggle competitions. But note, it IS against the rules to hand label anything from the test set.",
    "1229953": "Hello everyone,\n\nCan you tell me if doing manual annotation of training set is allowed? Doing this can be useful to train object detectors, for instance.",
    "1244704": "Hi, did you end up annotating the training data? If so, how do you go about doing it? Are there any automated way of doing it?",
    "1230717": ""
  }
}