{
  "id": 590973,
  "title": "Pre-Created Dataset Usage",
  "url": "/competitions/multi-class-object-detection-challenge/discussion/590973",
  "author_name": "",
  "post_date": "2025-07-24T13:13:24.447946700Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I had a few doubt about the 5 precreated dataset on the falcon competition page.<br>\n1) Could anyone please let me know are they unique data, apart from the kaggle data, or some samples from that is used to create kaggle competition data?<br>\n2) The test data in the precreated data, has labels as well, can we use this test-data, to train the models?<br>\n3) Lastly does the validation data in Base datatset also has real-world images, if not then I was thinking I can include it in model training ?</p>",
  "messages": [
    {
      "id": "3253329",
      "postDate": "07/24/2025 13:13:24",
      "content": "<p>I had a few doubt about the 5 precreated dataset on the falcon competition page.<br>\n1) Could anyone please let me know are they unique data, apart from the kaggle data, or some samples from that is used to create kaggle competition data?<br>\n2) The test data in the precreated data, has labels as well, can we use this test-data, to train the models?<br>\n3) Lastly does the validation data in Base datatset also has real-world images, if not then I was thinking I can include it in model training ?</p>",
      "rawMarkdown": "I had a few doubt about the 5 precreated dataset on the falcon competition page.\n1) Could anyone please let me know are they unique data, apart from the kaggle data, or some samples from that is used to create kaggle competition data?\n2) The test data in the precreated data, has labels as well, can we use this test-data, to train the models?\n3) Lastly does the validation data in Base datatset also has real-world images, if not then I was thinking I can include it in model training ?",
      "votes": null
    },
    {
      "id": "3253415",
      "postDate": "07/24/2025 16:20:18",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/yuvrajdhepe\" target=\"_blank\">@yuvrajdhepe</a>! I am a little confused by the \"5 precreated dataset\" part of your question. Do you mean the data in the data tab?<br>\n1.) I created all of the testing images on Falcon myself, and the real-world images were taken and annotated by me</p>\n<p>2.) There are some real-world images in the validation folder, those need to stay in validation and you can't use more real-world images than that for training or validation. The test data is for submitting predictions only, and we ask competition winners to provide an explanation of their strategy to check for cheating.</p>\n<p>3.) No, the real-world images can't be used for training. That is a rule violation. They need to be used for validation only, and do not add any more real images into your training workflow. However, you can augment the real-world images from the validation folder only.</p>\n<p>Let me know if you have any more questions!</p>",
      "rawMarkdown": "Hi @yuvrajdhepe! I am a little confused by the \"5 precreated dataset\" part of your question. Do you mean the data in the data tab?\n1.) I created all of the testing images on Falcon myself, and the real-world images were taken and annotated by me\n\n2.) There are some real-world images in the validation folder, those need to stay in validation and you can't use more real-world images than that for training or validation. The test data is for submitting predictions only, and we ask competition winners to provide an explanation of their strategy to check for cheating.\n\n3.) No, the real-world images can't be used for training. That is a rule violation. They need to be used for validation only, and do not add any more real images into your training workflow. However, you can augment the real-world images from the validation folder only.\n\nLet me know if you have any more questions!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3253415,
      "author_name": "rebekahduality",
      "author_url": "",
      "post_date": "07/24/2025 16:20:18",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/yuvrajdhepe\" target=\"_blank\">@yuvrajdhepe</a>! I am a little confused by the \"5 precreated dataset\" part of your question. Do you mean the data in the data tab?<br>\n1.) I created all of the testing images on Falcon myself, and the real-world images were taken and annotated by me</p>\n<p>2.) There are some real-world images in the validation folder, those need to stay in validation and you can't use more real-world images than that for training or validation. The test data is for submitting predictions only, and we ask competition winners to provide an explanation of their strategy to check for cheating.</p>\n<p>3.) No, the real-world images can't be used for training. That is a rule violation. They need to be used for validation only, and do not add any more real images into your training workflow. However, you can augment the real-world images from the validation folder only.</p>\n<p>Let me know if you have any more questions!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3253329": "I had a few doubt about the 5 precreated dataset on the falcon competition page.\n1) Could anyone please let me know are they unique data, apart from the kaggle data, or some samples from that is used to create kaggle competition data?\n2) The test data in the precreated data, has labels as well, can we use this test-data, to train the models?\n3) Lastly does the validation data in Base datatset also has real-world images, if not then I was thinking I can include it in model training ?",
    "3253415": "Hi @yuvrajdhepe! I am a little confused by the \"5 precreated dataset\" part of your question. Do you mean the data in the data tab?\n1.) I created all of the testing images on Falcon myself, and the real-world images were taken and annotated by me\n\n2.) There are some real-world images in the validation folder, those need to stay in validation and you can't use more real-world images than that for training or validation. The test data is for submitting predictions only, and we ask competition winners to provide an explanation of their strategy to check for cheating.\n\n3.) No, the real-world images can't be used for training. That is a rule violation. They need to be used for validation only, and do not add any more real images into your training workflow. However, you can augment the real-world images from the validation folder only.\n\nLet me know if you have any more questions!"
  },
  "source": "meta"
}