{
  "id": 651586,
  "title": "fogery images and authentic images",
  "url": "/competitions/recodai-luc-scientific-image-forgery-detection/discussion/651586",
  "author_name": "Japan",
  "post_date": "2025-12-04T09:13:41.646000",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>1) After inspecting more than 100+ images, I noticed inconsistencies between the authentic and forged samples. In several forged images, the masking seems poorly applied or incomplete — some regions appear to have been missed. (156.png,804.png,1237.png,1270.png, 2290.png,2121.png,……)</p>\n<p>2) For the supplemental_images, should we use the entire image as input, or are we expected to crop and focus only on the forged regions to detect/segment manipulation?</p>",
  "messages": [
    {
      "id": 3362247,
      "postDate": "2025-12-04T11:11:47.220Z",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/delliganesh\" target=\"_blank\">@delliganesh</a> ,</p>\n<p>Thank you for your questions and concerns.</p>\n<p>1) Regarding the mask inconsistencies: I reviewed the specific images you listed, but I could not find the issues regarding incomplete or poorly applied masks.</p>\n<p>Please be aware of two important factors that might explain what you are seeing:</p>\n<p>a) Multiple Channels: These images contain more than one copied region. Each distinct forgery instance is indicated in a different channel of the annotated masks. If you are visualizing them all together, they might look confusing.</p>\n<p>b-) Object Hiding: A forgery doesn't consist only of duplicating an object (e.g., a cell). It can also involve copying a part of the background to hide (or \"erase\") an object. These background copies are also annotated in the masks.</p>\n<p>If you still see issues after checking the individual channels and considering background hiding, please let me know specifically which coordinates look incomplete to you.</p>\n<p>2) Regarding supplemental_images: You are free to process them either by using the entire image or by cropping specific regions. However, regardless of the approach you choose, the final output must be remapped to the original image shape. Your submission must express the output mask in the same dimensions as the original image.</p>\n<p>Hope this helps!</p>",
      "rawMarkdown": "Hi @delliganesh ,\n\nThank you for your questions and concerns.\n\n1) Regarding the mask inconsistencies: I reviewed the specific images you listed, but I could not find the issues regarding incomplete or poorly applied masks.\n\nPlease be aware of two important factors that might explain what you are seeing:\n\na) Multiple Channels: These images contain more than one copied region. Each distinct forgery instance is indicated in a different channel of the annotated masks. If you are visualizing them all together, they might look confusing.\n\nb-) Object Hiding: A forgery doesn't consist only of duplicating an object (e.g., a cell). It can also involve copying a part of the background to hide (or \"erase\") an object. These background copies are also annotated in the masks.\n\nIf you still see issues after checking the individual channels and considering background hiding, please let me know specifically which coordinates look incomplete to you.\n\n2) Regarding supplemental_images: You are free to process them either by using the entire image or by cropping specific regions. However, regardless of the approach you choose, the final output must be remapped to the original image shape. Your submission must express the output mask in the same dimensions as the original image.\n\nHope this helps!",
      "votes": 1,
      "replies": [
        {
          "id": 3362342,
          "postDate": "2025-12-04T14:51:09.907Z",
          "rawMarkdown": "",
          "isDeleted": true
        },
        {
          "id": 3362392,
          "postDate": "2025-12-04T16:18:56.997Z",
          "content": "<p>by default i loaded with single channel , thats the issue.</p>",
          "rawMarkdown": "by default i loaded with single channel , thats the issue.",
          "votes": 1
        }
      ]
    },
    {
      "id": 3362177,
      "postDate": "2025-12-04T09:13:41.647Z",
      "content": "<p>1) After inspecting more than 100+ images, I noticed inconsistencies between the authentic and forged samples. In several forged images, the masking seems poorly applied or incomplete — some regions appear to have been missed. (156.png,804.png,1237.png,1270.png, 2290.png,2121.png,……)</p>\n<p>2) For the supplemental_images, should we use the entire image as input, or are we expected to crop and focus only on the forged regions to detect/segment manipulation?</p>",
      "rawMarkdown": "1) After inspecting more than 100+ images, I noticed inconsistencies between the authentic and forged samples. In several forged images, the masking seems poorly applied or incomplete — some regions appear to have been missed. (156.png,804.png,1237.png,1270.png, 2290.png,2121.png,......)\n\n2) For the supplemental_images, should we use the entire image as input, or are we expected to crop and focus only on the forged regions to detect/segment manipulation?",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 3362247,
      "author_name": "João Phillipe Cardenuto",
      "author_url": "",
      "post_date": "2025-12-04T11:11:47.220000",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/delliganesh\" target=\"_blank\">@delliganesh</a> ,</p>\n<p>Thank you for your questions and concerns.</p>\n<p>1) Regarding the mask inconsistencies: I reviewed the specific images you listed, but I could not find the issues regarding incomplete or poorly applied masks.</p>\n<p>Please be aware of two important factors that might explain what you are seeing:</p>\n<p>a) Multiple Channels: These images contain more than one copied region. Each distinct forgery instance is indicated in a different channel of the annotated masks. If you are visualizing them all together, they might look confusing.</p>\n<p>b-) Object Hiding: A forgery doesn't consist only of duplicating an object (e.g., a cell). It can also involve copying a part of the background to hide (or \"erase\") an object. These background copies are also annotated in the masks.</p>\n<p>If you still see issues after checking the individual channels and considering background hiding, please let me know specifically which coordinates look incomplete to you.</p>\n<p>2) Regarding supplemental_images: You are free to process them either by using the entire image or by cropping specific regions. However, regardless of the approach you choose, the final output must be remapped to the original image shape. Your submission must express the output mask in the same dimensions as the original image.</p>\n<p>Hope this helps!</p>",
      "votes": 1,
      "replies": [
        {
          "id": 3362342,
          "author_name": "",
          "author_url": "",
          "post_date": "2025-12-04T14:51:09.907000",
          "content": "",
          "votes": 0,
          "replies": []
        },
        {
          "id": 3362392,
          "author_name": "Japan",
          "author_url": "",
          "post_date": "2025-12-04T16:18:56.997000",
          "content": "<p>by default i loaded with single channel , thats the issue.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3362247": "Hi @delliganesh ,\n\nThank you for your questions and concerns.\n\n1) Regarding the mask inconsistencies: I reviewed the specific images you listed, but I could not find the issues regarding incomplete or poorly applied masks.\n\nPlease be aware of two important factors that might explain what you are seeing:\n\na) Multiple Channels: These images contain more than one copied region. Each distinct forgery instance is indicated in a different channel of the annotated masks. If you are visualizing them all together, they might look confusing.\n\nb-) Object Hiding: A forgery doesn't consist only of duplicating an object (e.g., a cell). It can also involve copying a part of the background to hide (or \"erase\") an object. These background copies are also annotated in the masks.\n\nIf you still see issues after checking the individual channels and considering background hiding, please let me know specifically which coordinates look incomplete to you.\n\n2) Regarding supplemental_images: You are free to process them either by using the entire image or by cropping specific regions. However, regardless of the approach you choose, the final output must be remapped to the original image shape. Your submission must express the output mask in the same dimensions as the original image.\n\nHope this helps!",
    "3362177": "1) After inspecting more than 100+ images, I noticed inconsistencies between the authentic and forged samples. In several forged images, the masking seems poorly applied or incomplete — some regions appear to have been missed. (156.png,804.png,1237.png,1270.png, 2290.png,2121.png,......)\n\n2) For the supplemental_images, should we use the entire image as input, or are we expected to crop and focus only on the forged regions to detect/segment manipulation?"
  }
}