{
  "id": 613293,
  "title": "Suggestion for Interactive Mode & Clarification on Test Set Path",
  "url": "/competitions/recodai-luc-scientific-image-forgery-detection/discussion/613293",
  "author_name": "Frank Morales",
  "post_date": "2025-10-25T15:46:55.782000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hello Kaggle Team and Competition Organizers,</p>\n<p>I’m working on the Recod.ai/LUC - Scientific Image Forgery Detection competition and have a suggestion to improve the development experience, as well as a question about the test set.</p>\n<p>Suggestion: Currently, interactive mode provides only 1 test image (45.png) in the /kaggle/input/recodai-luc-scientific-image-forgery-detection/test_images/ directory. This makes it challenging to debug pipelines effectively, especially for tasks involving segmentation, where edge cases (e.g., authentic vs. forged images) need testing. I suggest providing 3 test images in interactive mode instead of 1. This slight increase would allow developers to validate their models on a minimal but diverse set without risking overfitting, given the complete test set is ~1,100 images. It would streamline development and reduce reliance on commit-mode trial-and-error.</p>\n<p>Question: During commit mode, I’m only finding one image (45.png) at /kaggle/input/recodai-luc-scientific-image-forgery-detection/test_images/, despite the public leaderboard test set being described as ~1,100 images. What is the correct commit-time path for the complete test set? As a workaround, I’ve implemented a 5-row dummy submission (notebook link below), but I’d prefer to use the real data to ensure accurate forgery detection.</p>\n<p>Notebook Link: <a href=\"https://www.kaggle.com/code/frankmorales/forgery-agi\" target=\"_blank\">https://www.kaggle.com/code/frankmorales/forgery-agi</a></p>\n<p>I’d appreciate any feedback or guidance on the suggestion and the path issue. Thank you for being so supportive!</p>\n<p>Best,<br>\n<a href=\"https://www.kaggle.com/frankmorales\" target=\"_blank\">https://www.kaggle.com/frankmorales</a></p>",
  "messages": [
    {
      "id": 3306922,
      "postDate": "2025-10-25T15:46:55.783Z",
      "content": "<p>Hello Kaggle Team and Competition Organizers,</p>\n<p>I’m working on the Recod.ai/LUC - Scientific Image Forgery Detection competition and have a suggestion to improve the development experience, as well as a question about the test set.</p>\n<p>Suggestion: Currently, interactive mode provides only 1 test image (45.png) in the /kaggle/input/recodai-luc-scientific-image-forgery-detection/test_images/ directory. This makes it challenging to debug pipelines effectively, especially for tasks involving segmentation, where edge cases (e.g., authentic vs. forged images) need testing. I suggest providing 3 test images in interactive mode instead of 1. This slight increase would allow developers to validate their models on a minimal but diverse set without risking overfitting, given the complete test set is ~1,100 images. It would streamline development and reduce reliance on commit-mode trial-and-error.</p>\n<p>Question: During commit mode, I’m only finding one image (45.png) at /kaggle/input/recodai-luc-scientific-image-forgery-detection/test_images/, despite the public leaderboard test set being described as ~1,100 images. What is the correct commit-time path for the complete test set? As a workaround, I’ve implemented a 5-row dummy submission (notebook link below), but I’d prefer to use the real data to ensure accurate forgery detection.</p>\n<p>Notebook Link: <a href=\"https://www.kaggle.com/code/frankmorales/forgery-agi\" target=\"_blank\">https://www.kaggle.com/code/frankmorales/forgery-agi</a></p>\n<p>I’d appreciate any feedback or guidance on the suggestion and the path issue. Thank you for being so supportive!</p>\n<p>Best,<br>\n<a href=\"https://www.kaggle.com/frankmorales\" target=\"_blank\">https://www.kaggle.com/frankmorales</a></p>",
      "rawMarkdown": "Hello Kaggle Team and Competition Organizers,\n\nI’m working on the Recod.ai/LUC - Scientific Image Forgery Detection competition and have a suggestion to improve the development experience, as well as a question about the test set.\n\nSuggestion: Currently, interactive mode provides only 1 test image (45.png) in the /kaggle/input/recodai-luc-scientific-image-forgery-detection/test_images/ directory. This makes it challenging to debug pipelines effectively, especially for tasks involving segmentation, where edge cases (e.g., authentic vs. forged images) need testing. I suggest providing 3 test images in interactive mode instead of 1. This slight increase would allow developers to validate their models on a minimal but diverse set without risking overfitting, given the complete test set is ~1,100 images. It would streamline development and reduce reliance on commit-mode trial-and-error.\n\nQuestion: During commit mode, I’m only finding one image (45.png) at /kaggle/input/recodai-luc-scientific-image-forgery-detection/test_images/, despite the public leaderboard test set being described as ~1,100 images. What is the correct commit-time path for the complete test set? As a workaround, I’ve implemented a 5-row dummy submission (notebook link below), but I’d prefer to use the real data to ensure accurate forgery detection.\n\nNotebook Link: https://www.kaggle.com/code/frankmorales/forgery-agi\n\nI’d appreciate any feedback or guidance on the suggestion and the path issue. Thank you for being so supportive!\n\nBest,\nhttps://www.kaggle.com/frankmorales",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "3306922": "Hello Kaggle Team and Competition Organizers,\n\nI’m working on the Recod.ai/LUC - Scientific Image Forgery Detection competition and have a suggestion to improve the development experience, as well as a question about the test set.\n\nSuggestion: Currently, interactive mode provides only 1 test image (45.png) in the /kaggle/input/recodai-luc-scientific-image-forgery-detection/test_images/ directory. This makes it challenging to debug pipelines effectively, especially for tasks involving segmentation, where edge cases (e.g., authentic vs. forged images) need testing. I suggest providing 3 test images in interactive mode instead of 1. This slight increase would allow developers to validate their models on a minimal but diverse set without risking overfitting, given the complete test set is ~1,100 images. It would streamline development and reduce reliance on commit-mode trial-and-error.\n\nQuestion: During commit mode, I’m only finding one image (45.png) at /kaggle/input/recodai-luc-scientific-image-forgery-detection/test_images/, despite the public leaderboard test set being described as ~1,100 images. What is the correct commit-time path for the complete test set? As a workaround, I’ve implemented a 5-row dummy submission (notebook link below), but I’d prefer to use the real data to ensure accurate forgery detection.\n\nNotebook Link: https://www.kaggle.com/code/frankmorales/forgery-agi\n\nI’d appreciate any feedback or guidance on the suggestion and the path issue. Thank you for being so supportive!\n\nBest,\nhttps://www.kaggle.com/frankmorales"
  }
}