{ "id": 579617, "title": "🍿 Tips: Changing YOLO's default behavior", "url": "/competitions/byu-locating-bacterial-flagellar-motors-2025/discussion/579617", "author_name": "", "post_date": "2025-05-19T04:21:40.150260600Z", "votes": 24, "comment_count": 2, "views": 0, "content": "

Since YOLO is major tools for this competition, I share a simple trick to change default behavior of YOLO.

\n

Using the below snippet, we can dynamically changes YOLO's behavior.

\n
 typing  \n\n\n ():\n    \n      (obj, method_name):\n         AttributeError(\n            \n        )\n    bound = func.__get__(obj, obj.__class__)\n    (obj, method_name, bound)\n
\n

For example, the below notebook showcases how we can output featuremap and DFL prediction of YOLO.

\n

https://www.kaggle.com/code/tatamikenn/reverse-engineering-yolo

\n

\"\"
\n\"\"

", "messages": [ { "id": "3204928", "postDate": "05/19/2025 04:21:40", "content": "

Since YOLO is major tools for this competition, I share a simple trick to change default behavior of YOLO.

\n

Using the below snippet, we can dynamically changes YOLO's behavior.

\n
 typing  \n\n\n ():\n    \n      (obj, method_name):\n         AttributeError(\n            \n        )\n    bound = func.__get__(obj, obj.__class__)\n    (obj, method_name, bound)\n
\n

For example, the below notebook showcases how we can output featuremap and DFL prediction of YOLO.

\n

https://www.kaggle.com/code/tatamikenn/reverse-engineering-yolo

\n

\"\"
\n\"\"

", "rawMarkdown": "Since YOLO is major tools for this competition, I share a simple trick to change default behavior of YOLO.\n\nUsing the below snippet, we can dynamically changes YOLO's behavior.\n\n```python\nfrom typing import Callable\n\n\ndef replace_method(obj, method_name: str, func: Callable):\n \"\"\"\n Replace a method of an object with a new function.\n\n Args:\n obj: The object whose method is to be replaced.\n method_name (str): The name of the method to replace.\n func (FunctionType): The new function to set as the method.\n \"\"\"\n if not hasattr(obj, method_name):\n raise AttributeError(\n f\"{obj.__class__.__name__} has no method '{method_name}' to replace.\"\n )\n bound = func.__get__(obj, obj.__class__)\n setattr(obj, method_name, bound)\n```\n\nFor example, the below notebook showcases how we can output featuremap and DFL prediction of YOLO.\n\nhttps://www.kaggle.com/code/tatamikenn/reverse-engineering-yolo\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4910466%2Fcd0b0aa59734b8f19ec0c58103239d33%2FScreenshot%202025-05-19%20at%2013.19.04.png?generation=1747628443010165&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4910466%2F9c18cf4191e67e5e9d18b9558a810ff1%2FScreenshot%202025-05-19%20at%2012.57.07.png?generation=1747628465787381&alt=media)", "votes": null }, { "id": "3204932", "postDate": "05/19/2025 04:26:18", "content": "

You can also use this to extract feature.

\n
 ():\n    features = {}\n     ():\n         ():\n            features[name] = output\n         hook\n    model.model.model[idx].register_forward_hook(make_hook(feature_name))\n     features\n
", "rawMarkdown": "You can also use this to extract feature.\n\n```{python}\ndef get_feature_map(model, idx, feautre_name):\n features = {}\n def make_hook(name):\n def hook(module, input, output):\n features[name] = output\n return hook\n model.model.model[idx].register_forward_hook(make_hook(feature_name))\n return features\n```", "votes": null }, { "id": "3205943", "postDate": "05/20/2025 15:37:38", "content": "

Nice work 👍

", "rawMarkdown": "Nice work 👍", "votes": null } ], "comments": [ { "id": 3204932, "author_name": "tom99763", "author_url": "", "post_date": "05/19/2025 04:26:18", "content": "

You can also use this to extract feature.

\n
 ():\n    features = {}\n     ():\n         ():\n            features[name] = output\n         hook\n    model.model.model[idx].register_forward_hook(make_hook(feature_name))\n     features\n
", "votes": null, "replies": [] }, { "id": 3205943, "author_name": "aliraza1709", "author_url": "", "post_date": "05/20/2025 15:37:38", "content": "

Nice work 👍

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3204928": "Since YOLO is major tools for this competition, I share a simple trick to change default behavior of YOLO.\n\nUsing the below snippet, we can dynamically changes YOLO's behavior.\n\n```python\nfrom typing import Callable\n\n\ndef replace_method(obj, method_name: str, func: Callable):\n \"\"\"\n Replace a method of an object with a new function.\n\n Args:\n obj: The object whose method is to be replaced.\n method_name (str): The name of the method to replace.\n func (FunctionType): The new function to set as the method.\n \"\"\"\n if not hasattr(obj, method_name):\n raise AttributeError(\n f\"{obj.__class__.__name__} has no method '{method_name}' to replace.\"\n )\n bound = func.__get__(obj, obj.__class__)\n setattr(obj, method_name, bound)\n```\n\nFor example, the below notebook showcases how we can output featuremap and DFL prediction of YOLO.\n\nhttps://www.kaggle.com/code/tatamikenn/reverse-engineering-yolo\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4910466%2Fcd0b0aa59734b8f19ec0c58103239d33%2FScreenshot%202025-05-19%20at%2013.19.04.png?generation=1747628443010165&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F4910466%2F9c18cf4191e67e5e9d18b9558a810ff1%2FScreenshot%202025-05-19%20at%2012.57.07.png?generation=1747628465787381&alt=media)", "3204932": "You can also use this to extract feature.\n\n```{python}\ndef get_feature_map(model, idx, feautre_name):\n features = {}\n def make_hook(name):\n def hook(module, input, output):\n features[name] = output\n return hook\n model.model.model[idx].register_forward_hook(make_hook(feature_name))\n return features\n```", "3205943": "Nice work 👍" }, "source": "meta" }