Hi everyone,
\nI’m currently participating in this competition and I have a question about the use of pretrained models or pretrained weights.
\nAccording to the competition rules, I understand that external data and pretrained models might be subject to certain restrictions. However, I’d like to confirm whether the following is allowed:
\nUsing pretrained weights from models trained on ImageNet / COCO / other public datasets (e.g., ResNet, EfficientNet, etc.)
\nFine-tuning those pretrained models on the official competition data only
\nIn other words, I’m not planning to use any additional external data — just starting from a pretrained backbone.
\nCould anyone (or a competition host / moderator) please confirm if this is acceptable under the current competition rules?
\nThanks in advance for your help! 🙏
", "messages": [ { "id": "3308904", "postDate": "10/30/2025 14:19:31", "content": "Hi everyone,
\nI’m currently participating in this competition and I have a question about the use of pretrained models or pretrained weights.
\nAccording to the competition rules, I understand that external data and pretrained models might be subject to certain restrictions. However, I’d like to confirm whether the following is allowed:
\nUsing pretrained weights from models trained on ImageNet / COCO / other public datasets (e.g., ResNet, EfficientNet, etc.)
\nFine-tuning those pretrained models on the official competition data only
\nIn other words, I’m not planning to use any additional external data — just starting from a pretrained backbone.
\nCould anyone (or a competition host / moderator) please confirm if this is acceptable under the current competition rules?
\nThanks in advance for your help! 🙏
", "rawMarkdown": "Hi everyone,\n\nI’m currently participating in this competition and I have a question about the use of pretrained models or pretrained weights.\n\nAccording to the competition rules, I understand that external data and pretrained models might be subject to certain restrictions. However, I’d like to confirm whether the following is allowed:\n\nUsing pretrained weights from models trained on ImageNet / COCO / other public datasets (e.g., ResNet, EfficientNet, etc.)\n\nFine-tuning those pretrained models on the official competition data only\n\nIn other words, I’m not planning to use any additional external data — just starting from a pretrained backbone.\n\nCould anyone (or a competition host / moderator) please confirm if this is acceptable under the current competition rules?\n\nThanks in advance for your help! 🙏", "votes": null }, { "id": "3308921", "postDate": "10/30/2025 14:52:25", "content": "Hi,
\nFreely and publicly available external data is allowed, including pre-trained models. So, it seems that there is no problem following your idea.
\nDouble-check the competition's code requirements for more information on this.
\n", "rawMarkdown": "Hi,\n\nFreely and publicly available external data is allowed, including pre-trained models. So, it seems that there is no problem following your idea.\n\nDouble-check the competition's code requirements for more information on this.\n\nwww.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/overview/code-requirements", "votes": null }, { "id": "3308922", "postDate": "10/30/2025 14:53:44", "content": "You can do it
", "rawMarkdown": "You can do it", "votes": null }, { "id": "3308927", "postDate": "10/30/2025 14:56:53", "content": "Hi, thank you for clarifying!
\nJust to confirm — if I train a model locally and then upload my own trained weights (not publicly available pretrained ones) for inference during submission, is that allowed?\nIn other words, can I submit inference-only code without including the training process in my notebook?
", "rawMarkdown": "Hi, thank you for clarifying!\n\nJust to confirm — if I train a model locally and then upload my own trained weights (not publicly available pretrained ones) for inference during submission, is that allowed?\nIn other words, can I submit inference-only code without including the training process in my notebook?", "votes": null }, { "id": "3308982", "postDate": "10/30/2025 16:38:58", "content": "As long as the data you are using is accessible to all participants (i.e., others should be able to reproduce your data methodology) and your method is made open-source after the competition, there is no issue.
\nPlease remember two key points from the competition rules:
\nhttps://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/rules
\nHope that helps :) Good luck and have fun during the competition!
", "rawMarkdown": "As long as the data you are using is accessible to all participants (i.e., others should be able to reproduce your data methodology) and your method is made open-source after the competition, there is no issue.\n\nPlease remember two key points from the competition rules:\n1. Any external data you use must be equally accessible to all other participants (See Section 6 of the rules).\n2. Winning solutions will be licensed as open-source, and you may be required to detail your methodology to allow others to reproduce your technique. (See Section 5 of the rules).\n\nhttps://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/rules\n\nHope that helps :) Good luck and have fun during the competition!", "votes": null } ], "comments": [ { "id": 3308921, "author_name": "joophillipecardenuto", "author_url": "", "post_date": "10/30/2025 14:52:25", "content": "Hi,
\nFreely and publicly available external data is allowed, including pre-trained models. So, it seems that there is no problem following your idea.
\nDouble-check the competition's code requirements for more information on this.
\n", "votes": null, "replies": [ { "id": 3308927, "author_name": "", "author_url": "", "post_date": "10/30/2025 14:56:53", "content": "Hi, thank you for clarifying!
\nJust to confirm — if I train a model locally and then upload my own trained weights (not publicly available pretrained ones) for inference during submission, is that allowed?\nIn other words, can I submit inference-only code without including the training process in my notebook?
", "votes": null, "replies": [ { "id": 3308982, "author_name": "joophillipecardenuto", "author_url": "", "post_date": "10/30/2025 16:38:58", "content": "As long as the data you are using is accessible to all participants (i.e., others should be able to reproduce your data methodology) and your method is made open-source after the competition, there is no issue.
\nPlease remember two key points from the competition rules:
\nhttps://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/rules
\nHope that helps :) Good luck and have fun during the competition!
", "votes": null, "replies": [] } ] } ] }, { "id": 3308922, "author_name": "qifeihhh666", "author_url": "", "post_date": "10/30/2025 14:53:44", "content": "You can do it
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3308904": "Hi everyone,\n\nI’m currently participating in this competition and I have a question about the use of pretrained models or pretrained weights.\n\nAccording to the competition rules, I understand that external data and pretrained models might be subject to certain restrictions. However, I’d like to confirm whether the following is allowed:\n\nUsing pretrained weights from models trained on ImageNet / COCO / other public datasets (e.g., ResNet, EfficientNet, etc.)\n\nFine-tuning those pretrained models on the official competition data only\n\nIn other words, I’m not planning to use any additional external data — just starting from a pretrained backbone.\n\nCould anyone (or a competition host / moderator) please confirm if this is acceptable under the current competition rules?\n\nThanks in advance for your help! 🙏", "3308921": "Hi,\n\nFreely and publicly available external data is allowed, including pre-trained models. So, it seems that there is no problem following your idea.\n\nDouble-check the competition's code requirements for more information on this.\n\nwww.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/overview/code-requirements", "3308922": "You can do it", "3308927": "Hi, thank you for clarifying!\n\nJust to confirm — if I train a model locally and then upload my own trained weights (not publicly available pretrained ones) for inference during submission, is that allowed?\nIn other words, can I submit inference-only code without including the training process in my notebook?", "3308982": "As long as the data you are using is accessible to all participants (i.e., others should be able to reproduce your data methodology) and your method is made open-source after the competition, there is no issue.\n\nPlease remember two key points from the competition rules:\n1. Any external data you use must be equally accessible to all other participants (See Section 6 of the rules).\n2. Winning solutions will be licensed as open-source, and you may be required to detail your methodology to allow others to reproduce your technique. (See Section 5 of the rules).\n\nhttps://www.kaggle.com/competitions/recodai-luc-scientific-image-forgery-detection/rules\n\nHope that helps :) Good luck and have fun during the competition!" }, "source": "meta" }