{
  "id": 313077,
  "title": "Using test images for training is OK?",
  "url": "/competitions/happy-whale-and-dolphin/discussion/313077",
  "author_name": "",
  "post_date": "2022-03-15T14:25:16.054023500Z",
  "votes": 2,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hello, I'm a kaggle beginner and this is my first competition. Here, I have a very basic question about rules.</p>\n<p>Although correct labels for test images are not given, it may be possible to use the unlabeled test images during training to boost the performance using semi-supervised learning, self-supervised learning and so on. Is it OK to use the test images in such way? I could not find any descriptions about that on the competition page.</p>",
  "messages": [
    {
      "id": "1723567",
      "postDate": "03/15/2022 14:25:16",
      "content": "<p>Hello, I'm a kaggle beginner and this is my first competition. Here, I have a very basic question about rules.</p>\n<p>Although correct labels for test images are not given, it may be possible to use the unlabeled test images during training to boost the performance using semi-supervised learning, self-supervised learning and so on. Is it OK to use the test images in such way? I could not find any descriptions about that on the competition page.</p>",
      "rawMarkdown": "Hello, I'm a kaggle beginner and this is my first competition. Here, I have a very basic question about rules.\n\nAlthough correct labels for test images are not given, it may be possible to use the unlabeled test images during training to boost the performance using semi-supervised learning, self-supervised learning and so on. Is it OK to use the test images in such way? I could not find any descriptions about that on the competition page.",
      "votes": null
    },
    {
      "id": "1723606",
      "postDate": "03/15/2022 14:55:17",
      "content": "<p>Yes - you can.  You cannot hand label test images and use them - it's very common to add high probability test images to the training.</p>",
      "rawMarkdown": "Yes - you can.  You cannot hand label test images and use them - it's very common to add high probability test images to the training.",
      "votes": null
    },
    {
      "id": "1723898",
      "postDate": "03/15/2022 20:52:44",
      "content": "<p>Thank you! It's really important tips.</p>",
      "rawMarkdown": "Thank you! It's really important tips.",
      "votes": null
    },
    {
      "id": "1724748",
      "postDate": "03/16/2022 13:29:02",
      "content": "<p>To add to what Jimmy mentioned: The technique is known as \"Pseudo-Labelling\" if you would like to checkout any tutorials.</p>",
      "rawMarkdown": "To add to what Jimmy mentioned: The technique is known as \"Pseudo-Labelling\" if you would like to checkout any tutorials.",
      "votes": null
    },
    {
      "id": "1725858",
      "postDate": "03/17/2022 13:38:07",
      "content": "<p>Thank you. I'll try that. But before I do, I want to know the impact of \"pseudo-labellling\" test images on THIS competition. So, if you know, could you tell me any codes or discussions relating to \"pseudo-labellling?\"</p>",
      "rawMarkdown": "Thank you. I'll try that. But before I do, I want to know the impact of \"pseudo-labellling\" test images on THIS competition. So, if you know, could you tell me any codes or discussions relating to \"pseudo-labellling?\"",
      "votes": null
    },
    {
      "id": "1727835",
      "postDate": "03/18/2022 11:17:42",
      "content": "<p>I'm not sure if it has been discussed at all in this competition. However, if you're looking for examples from older ones, Chris Deotte had shared <a href=\"https://www.kaggle.com/cdeotte/pseudo-labeling-qda-0-969\" target=\"_blank\">this</a> awesome kernel. </p>",
      "rawMarkdown": "I'm not sure if it has been discussed at all in this competition. However, if you're looking for examples from older ones, Chris Deotte had shared [this](https://www.kaggle.com/cdeotte/pseudo-labeling-qda-0-969) awesome kernel.",
      "votes": null
    },
    {
      "id": "1728881",
      "postDate": "03/19/2022 11:26:02",
      "content": "<p>I appreciate your kindness. I like Kaggle community.</p>",
      "rawMarkdown": "I appreciate your kindness. I like Kaggle community.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1723606,
      "author_name": "pcjimmmy",
      "author_url": "",
      "post_date": "03/15/2022 14:55:17",
      "content": "<p>Yes - you can.  You cannot hand label test images and use them - it's very common to add high probability test images to the training.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1723898,
          "author_name": "simeji",
          "author_url": "",
          "post_date": "03/15/2022 20:52:44",
          "content": "<p>Thank you! It's really important tips.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1724748,
          "author_name": "init27",
          "author_url": "",
          "post_date": "03/16/2022 13:29:02",
          "content": "<p>To add to what Jimmy mentioned: The technique is known as \"Pseudo-Labelling\" if you would like to checkout any tutorials.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1725858,
          "author_name": "simeji",
          "author_url": "",
          "post_date": "03/17/2022 13:38:07",
          "content": "<p>Thank you. I'll try that. But before I do, I want to know the impact of \"pseudo-labellling\" test images on THIS competition. So, if you know, could you tell me any codes or discussions relating to \"pseudo-labellling?\"</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1727835,
          "author_name": "init27",
          "author_url": "",
          "post_date": "03/18/2022 11:17:42",
          "content": "<p>I'm not sure if it has been discussed at all in this competition. However, if you're looking for examples from older ones, Chris Deotte had shared <a href=\"https://www.kaggle.com/cdeotte/pseudo-labeling-qda-0-969\" target=\"_blank\">this</a> awesome kernel. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1728881,
          "author_name": "simeji",
          "author_url": "",
          "post_date": "03/19/2022 11:26:02",
          "content": "<p>I appreciate your kindness. I like Kaggle community.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1723567": "Hello, I'm a kaggle beginner and this is my first competition. Here, I have a very basic question about rules.\n\nAlthough correct labels for test images are not given, it may be possible to use the unlabeled test images during training to boost the performance using semi-supervised learning, self-supervised learning and so on. Is it OK to use the test images in such way? I could not find any descriptions about that on the competition page.",
    "1723606": "Yes - you can.  You cannot hand label test images and use them - it's very common to add high probability test images to the training.",
    "1723898": "Thank you! It's really important tips.",
    "1724748": "To add to what Jimmy mentioned: The technique is known as \"Pseudo-Labelling\" if you would like to checkout any tutorials.",
    "1725858": "Thank you. I'll try that. But before I do, I want to know the impact of \"pseudo-labellling\" test images on THIS competition. So, if you know, could you tell me any codes or discussions relating to \"pseudo-labellling?\"",
    "1727835": "I'm not sure if it has been discussed at all in this competition. However, if you're looking for examples from older ones, Chris Deotte had shared [this](https://www.kaggle.com/cdeotte/pseudo-labeling-qda-0-969) awesome kernel.",
    "1728881": "I appreciate your kindness. I like Kaggle community."
  },
  "source": "meta"
}