{
  "id": 664624,
  "title": "Clarification on Copy–Move Forgery Variants and Transformations in the Dataset",
  "url": "/competitions/recodai-luc-scientific-image-forgery-detection/discussion/664624",
  "author_name": "",
  "post_date": "2025-12-26T19:39:06.713998400Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>In the context of this competition, could you please clarify the expected characteristics of copy–move forgeries present in the dataset?</p>\n<p>Specifically:\n– Can a single image contain multiple copied regions, or is each forged image limited to one copy–move instance?\n– Are copied regions always duplicated exactly, or do they include transformations such as rotation, scaling, flipping, elastic deformation, or intensity/contrast changes?\n– Are copied regions expected to preserve their original shape, or can they be warped or partially blended into surrounding structures?\n– Are forgeries restricted to intra-image copy–move only, or can they involve correlated but non-identical patterns?</p>\n<p>Clarification on these points would greatly help in designing appropriate augmentation strategies and evaluation assumptions. <a href=\"https://www.kaggle.com/joophillipecardenuto\" target=\"_blank\">@joophillipecardenuto</a> </p>",
  "messages": [
    {
      "id": "3382186",
      "postDate": "12/26/2025 19:39:06",
      "content": "<p>In the context of this competition, could you please clarify the expected characteristics of copy–move forgeries present in the dataset?</p>\n<p>Specifically:\n– Can a single image contain multiple copied regions, or is each forged image limited to one copy–move instance?\n– Are copied regions always duplicated exactly, or do they include transformations such as rotation, scaling, flipping, elastic deformation, or intensity/contrast changes?\n– Are copied regions expected to preserve their original shape, or can they be warped or partially blended into surrounding structures?\n– Are forgeries restricted to intra-image copy–move only, or can they involve correlated but non-identical patterns?</p>\n<p>Clarification on these points would greatly help in designing appropriate augmentation strategies and evaluation assumptions. <a href=\"https://www.kaggle.com/joophillipecardenuto\" target=\"_blank\">@joophillipecardenuto</a> </p>",
      "rawMarkdown": "In the context of this competition, could you please clarify the expected characteristics of copy–move forgeries present in the dataset?\n\nSpecifically:\n– Can a single image contain multiple copied regions, or is each forged image limited to one copy–move instance?\n– Are copied regions always duplicated exactly, or do they include transformations such as rotation, scaling, flipping, elastic deformation, or intensity/contrast changes?\n– Are copied regions expected to preserve their original shape, or can they be warped or partially blended into surrounding structures?\n– Are forgeries restricted to intra-image copy–move only, or can they involve correlated but non-identical patterns?\n\nClarification on these points would greatly help in designing appropriate augmentation strategies and evaluation assumptions. @joophillipecardenuto",
      "votes": null
    },
    {
      "id": "3382237",
      "postDate": "12/27/2025 02:33:10",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/jamalsaeedi\" target=\"_blank\">@jamalsaeedi</a>!</p>\n<p>Here are some answers:</p>\n<ol>\n<li><p>Can a single image contain multiple copied regions? Yes, a single image can contain multiple copied regions and multiple distinct forgery instances.</p></li>\n<li><p>Are copied regions always duplicated exactly? No. The copied regions can include transformations such as the ones you mentioned (rotation, scaling, flipping, etc.).</p></li>\n<li><p>Are copied regions expected to preserve their original shape? Copied regions can be distorted, often due to scaling or deformation. Generally, you should expect that the source region is transformed by a homography transformation.</p></li>\n<li><p>Are forgeries restricted to intra-image copy–move only? Due to the transformations mentioned above, as well as the addition of noise or compression artifacts, you might find non-identical regions. However, all source and copied regions are correlated and can usually be visually identified (though this is often difficult for non-experts).</p></li>\n</ol>\n<p>You may find more information in this post: <a href=\"https://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/discussion/614069\" target=\"_blank\">https://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/discussion/614069</a></p>\n<p>Or in this article: <a href=\"https://link.springer.com/article/10.1007/s11948-022-00391-4\" target=\"_blank\">https://link.springer.com/article/10.1007/s11948-022-00391-4</a></p>\n<p>Hope that helps :)</p>",
      "rawMarkdown": "Hi @jamalsaeedi!\n\nHere are some answers:\n\n1. Can a single image contain multiple copied regions? Yes, a single image can contain multiple copied regions and multiple distinct forgery instances.\n\n2. Are copied regions always duplicated exactly? No. The copied regions can include transformations such as the ones you mentioned (rotation, scaling, flipping, etc.).\n\n3. Are copied regions expected to preserve their original shape? Copied regions can be distorted, often due to scaling or deformation. Generally, you should expect that the source region is transformed by a homography transformation.\n\n4. Are forgeries restricted to intra-image copy–move only? Due to the transformations mentioned above, as well as the addition of noise or compression artifacts, you might find non-identical regions. However, all source and copied regions are correlated and can usually be visually identified (though this is often difficult for non-experts).\n\nYou may find more information in this post: [https://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/discussion/614069](https://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/discussion/614069)\n\nOr in this article: [https://link.springer.com/article/10.1007/s11948-022-00391-4](https://link.springer.com/article/10.1007/s11948-022-00391-4)\n\nHope that helps :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3382237,
      "author_name": "joophillipecardenuto",
      "author_url": "",
      "post_date": "12/27/2025 02:33:10",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/jamalsaeedi\" target=\"_blank\">@jamalsaeedi</a>!</p>\n<p>Here are some answers:</p>\n<ol>\n<li><p>Can a single image contain multiple copied regions? Yes, a single image can contain multiple copied regions and multiple distinct forgery instances.</p></li>\n<li><p>Are copied regions always duplicated exactly? No. The copied regions can include transformations such as the ones you mentioned (rotation, scaling, flipping, etc.).</p></li>\n<li><p>Are copied regions expected to preserve their original shape? Copied regions can be distorted, often due to scaling or deformation. Generally, you should expect that the source region is transformed by a homography transformation.</p></li>\n<li><p>Are forgeries restricted to intra-image copy–move only? Due to the transformations mentioned above, as well as the addition of noise or compression artifacts, you might find non-identical regions. However, all source and copied regions are correlated and can usually be visually identified (though this is often difficult for non-experts).</p></li>\n</ol>\n<p>You may find more information in this post: <a href=\"https://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/discussion/614069\" target=\"_blank\">https://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/discussion/614069</a></p>\n<p>Or in this article: <a href=\"https://link.springer.com/article/10.1007/s11948-022-00391-4\" target=\"_blank\">https://link.springer.com/article/10.1007/s11948-022-00391-4</a></p>\n<p>Hope that helps :)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3382186": "In the context of this competition, could you please clarify the expected characteristics of copy–move forgeries present in the dataset?\n\nSpecifically:\n– Can a single image contain multiple copied regions, or is each forged image limited to one copy–move instance?\n– Are copied regions always duplicated exactly, or do they include transformations such as rotation, scaling, flipping, elastic deformation, or intensity/contrast changes?\n– Are copied regions expected to preserve their original shape, or can they be warped or partially blended into surrounding structures?\n– Are forgeries restricted to intra-image copy–move only, or can they involve correlated but non-identical patterns?\n\nClarification on these points would greatly help in designing appropriate augmentation strategies and evaluation assumptions. @joophillipecardenuto",
    "3382237": "Hi @jamalsaeedi!\n\nHere are some answers:\n\n1. Can a single image contain multiple copied regions? Yes, a single image can contain multiple copied regions and multiple distinct forgery instances.\n\n2. Are copied regions always duplicated exactly? No. The copied regions can include transformations such as the ones you mentioned (rotation, scaling, flipping, etc.).\n\n3. Are copied regions expected to preserve their original shape? Copied regions can be distorted, often due to scaling or deformation. Generally, you should expect that the source region is transformed by a homography transformation.\n\n4. Are forgeries restricted to intra-image copy–move only? Due to the transformations mentioned above, as well as the addition of noise or compression artifacts, you might find non-identical regions. However, all source and copied regions are correlated and can usually be visually identified (though this is often difficult for non-experts).\n\nYou may find more information in this post: [https://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/discussion/614069](https://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/discussion/614069)\n\nOr in this article: [https://link.springer.com/article/10.1007/s11948-022-00391-4](https://link.springer.com/article/10.1007/s11948-022-00391-4)\n\nHope that helps :)"
  },
  "source": "meta"
}