{
  "id": 78459,
  "title": "Can I use test images to train the model?",
  "url": "/competitions/humpback-whale-identification/discussion/78459",
  "author_name": "",
  "post_date": "2019-01-24T04:29:36.433387Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Can I use test images to train the model?\nIs it against the rules?</p>",
  "messages": [
    {
      "id": "460607",
      "postDate": "01/24/2019 04:29:36",
      "content": "<p>Can I use test images to train the model?\nIs it against the rules?</p>",
      "rawMarkdown": "Can I use test images to train the model?\nIs it against the rules?",
      "votes": null
    },
    {
      "id": "460673",
      "postDate": "01/24/2019 08:01:51",
      "content": "<p>Hand labeling is against of the rule. So if you will use your model to assign labels to the test images it is ok, but you will definitely overfitt your solution by doing this. I warn you</p>",
      "rawMarkdown": "Hand labeling is against of the rule. So if you will use your model to assign labels to the test images it is ok, but you will definitely overfitt your solution by doing this. I warn you",
      "votes": null
    },
    {
      "id": "460682",
      "postDate": "01/24/2019 08:34:13",
      "content": "<p>Thank you. </p>\n\n<p>I just want to train a model to find where is the tail. (segmentation?)</p>",
      "rawMarkdown": "Thank you. \n\nI just want to train a model to find where is the tail. (segmentation?)",
      "votes": null
    },
    {
      "id": "461092",
      "postDate": "01/25/2019 07:45:40",
      "content": "<p>As the rules state, \"Entrants may re-annotate images in the training set, but may not hand-label predictions, including having human observers rate and evaluate the test data set.\", but I think labeling bboxes/segmentation is fine (though I don't think it makes a lot of sense to label all the training and test set for this)</p>",
      "rawMarkdown": "As the rules state, \"Entrants may re-annotate images in the training set, but may not hand-label predictions, including having human observers rate and evaluate the test data set.\", but I think labeling bboxes/segmentation is fine (though I don't think it makes a lot of sense to label all the training and test set for this)",
      "votes": null
    },
    {
      "id": "466939",
      "postDate": "02/06/2019 08:34:01",
      "content": "<p>what about doing manual segmentation for each id? </p>",
      "rawMarkdown": "what about doing manual segmentation for each id?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 460673,
      "author_name": "pinullmezon",
      "author_url": "",
      "post_date": "01/24/2019 08:01:51",
      "content": "<p>Hand labeling is against of the rule. So if you will use your model to assign labels to the test images it is ok, but you will definitely overfitt your solution by doing this. I warn you</p>",
      "votes": null,
      "replies": [
        {
          "id": 460682,
          "author_name": "wuliaokaola",
          "author_url": "",
          "post_date": "01/24/2019 08:34:13",
          "content": "<p>Thank you. </p>\n\n<p>I just want to train a model to find where is the tail. (segmentation?)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 461092,
          "author_name": "devvindan",
          "author_url": "",
          "post_date": "01/25/2019 07:45:40",
          "content": "<p>As the rules state, \"Entrants may re-annotate images in the training set, but may not hand-label predictions, including having human observers rate and evaluate the test data set.\", but I think labeling bboxes/segmentation is fine (though I don't think it makes a lot of sense to label all the training and test set for this)</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 466939,
          "author_name": "kathyppinos",
          "author_url": "",
          "post_date": "02/06/2019 08:34:01",
          "content": "<p>what about doing manual segmentation for each id? </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "460607": "Can I use test images to train the model?\nIs it against the rules?",
    "460673": "Hand labeling is against of the rule. So if you will use your model to assign labels to the test images it is ok, but you will definitely overfitt your solution by doing this. I warn you",
    "460682": "Thank you. \n\nI just want to train a model to find where is the tail. (segmentation?)",
    "461092": "As the rules state, \"Entrants may re-annotate images in the training set, but may not hand-label predictions, including having human observers rate and evaluate the test data set.\", but I think labeling bboxes/segmentation is fine (though I don't think it makes a lot of sense to label all the training and test set for this)",
    "466939": "what about doing manual segmentation for each id?"
  },
  "source": "meta"
}