{ "id": 572003, "title": "Yolo training details", "url": "/competitions/byu-locating-bacterial-flagellar-motors-2025/discussion/572003", "author_name": "", "post_date": "2025-04-07T04:34:01.483159500Z", "votes": 5, "comment_count": 6, "views": 0, "content": "

Can anyone share the training details of yolo? Mine stops after 37 epochs. I think I am missing something out

", "messages": [ { "id": "3172676", "postDate": "04/07/2025 04:34:01", "content": "

Can anyone share the training details of yolo? Mine stops after 37 epochs. I think I am missing something out

", "rawMarkdown": "Can anyone share the training details of yolo? Mine stops after 37 epochs. I think I am missing something out", "votes": null }, { "id": "3172732", "postDate": "04/07/2025 05:51:03", "content": "

Several notebooks already share the Yolo training model.\nBase line by Andrew Darley host of this competition\nAnother one Improved training From Mehdi Sharifi based on the base line\nHope it helps.

", "rawMarkdown": "Several notebooks already share the Yolo training model.\n[Base line by Andrew Darley host of this competition](https://www.kaggle.com/code/andrewjdarley/train-yolo)\nAnother one [Improved training From Mehdi Sharifi based on the base line](https://www.kaggle.com/code/sharifi76/eda-visualization-yolov8)\nHope it helps.", "votes": null }, { "id": "3173178", "postDate": "04/07/2025 16:06:42", "content": "

I'd take a look at the notebooks mentioned by @mathieuduverne as well as this one which has been well used in this competition. Also check out my colleague Jackson's notebook.

", "rawMarkdown": "I'd take a look at the notebooks mentioned by @mathieuduverne as well as [this one](https://www.kaggle.com/code/engadamalmohammedi/byu-lb-0-642-yolov8l-yolov11-ensemble) which has been well used in this competition. Also check out my colleague Jackson's [notebook](https://www.kaggle.com/code/jacksonpond/demo-notebook).", "votes": null }, { "id": "3173436", "postDate": "04/07/2025 22:42:54", "content": "

YOLO has an early stopping function in its arguments, so if training stops before reaching the set number of epochs, that is the cause.

\n
\n\n=, \n
", "rawMarkdown": "YOLO has an early stopping function in its arguments, so if training stops before reaching the set number of epochs, that is the cause.\n```\n# Number of epochs to wait without improvement in validation metrics before early stopping the training. \n# Helps prevent overfitting by stopping training when performance plateaus.\npatience=5, \n```", "votes": null }, { "id": "3173531", "postDate": "04/08/2025 03:43:02", "content": "

@hideyukizushi I was trying to say that I am not seeing any improvement after 30-35 epochs, using adamW optimizer lr- 0.0005
\nFine tuning after 100 epochs gave 0.57 lb
\nShould I perform some sort of preprocessing before training for improvement?

", "rawMarkdown": "hideyukizushi I was trying to say that I am not seeing any improvement after 30-35 epochs, using adamW optimizer lr- 0.0005 \nFine tuning after 100 epochs gave 0.57 lb\nShould I perform some sort of preprocessing before training for improvement?", "votes": null }, { "id": "3173721", "postDate": "04/08/2025 08:50:39", "content": "

Yes, did you see this discussion? Preprocessing approach by Nikolai Prudnikov
\nSeveral pre-processing techniques such as augmentation, denoising, normalization,etc can be tried out.

", "rawMarkdown": "Yes, did you see this discussion? [Preprocessing approach by Nikolai Prudnikov](https://www.kaggle.com/competitions/byu-locating-bacterial-flagellar-motors-2025/discussion/569996)\nSeveral pre-processing techniques such as augmentation, denoising, normalization,etc can be tried out.", "votes": null }, { "id": "3173728", "postDate": "04/08/2025 08:59:11", "content": "

Thanks, I discovered the @jacksonpond Notebook

", "rawMarkdown": "Thanks, I discovered the @jacksonpond Notebook", "votes": null } ], "comments": [ { "id": 3172732, "author_name": "mathieuduverne", "author_url": "", "post_date": "04/07/2025 05:51:03", "content": "

Several notebooks already share the Yolo training model.\nBase line by Andrew Darley host of this competition\nAnother one Improved training From Mehdi Sharifi based on the base line\nHope it helps.

", "votes": null, "replies": [] }, { "id": 3173178, "author_name": "andrewjdarley", "author_url": "", "post_date": "04/07/2025 16:06:42", "content": "

I'd take a look at the notebooks mentioned by @mathieuduverne as well as this one which has been well used in this competition. Also check out my colleague Jackson's notebook.

", "votes": null, "replies": [ { "id": 3173728, "author_name": "mathieuduverne", "author_url": "", "post_date": "04/08/2025 08:59:11", "content": "

Thanks, I discovered the @jacksonpond Notebook

", "votes": null, "replies": [] } ] }, { "id": 3173436, "author_name": "hideyukizushi", "author_url": "", "post_date": "04/07/2025 22:42:54", "content": "

YOLO has an early stopping function in its arguments, so if training stops before reaching the set number of epochs, that is the cause.

\n
\n\n=, \n
", "votes": null, "replies": [ { "id": 3173531, "author_name": "aksh1t", "author_url": "", "post_date": "04/08/2025 03:43:02", "content": "

@hideyukizushi I was trying to say that I am not seeing any improvement after 30-35 epochs, using adamW optimizer lr- 0.0005
\nFine tuning after 100 epochs gave 0.57 lb
\nShould I perform some sort of preprocessing before training for improvement?

", "votes": null, "replies": [ { "id": 3173721, "author_name": "mathieuduverne", "author_url": "", "post_date": "04/08/2025 08:50:39", "content": "

Yes, did you see this discussion? Preprocessing approach by Nikolai Prudnikov
\nSeveral pre-processing techniques such as augmentation, denoising, normalization,etc can be tried out.

", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3172676": "Can anyone share the training details of yolo? Mine stops after 37 epochs. I think I am missing something out", "3172732": "Several notebooks already share the Yolo training model.\n[Base line by Andrew Darley host of this competition](https://www.kaggle.com/code/andrewjdarley/train-yolo)\nAnother one [Improved training From Mehdi Sharifi based on the base line](https://www.kaggle.com/code/sharifi76/eda-visualization-yolov8)\nHope it helps.", "3173178": "I'd take a look at the notebooks mentioned by @mathieuduverne as well as [this one](https://www.kaggle.com/code/engadamalmohammedi/byu-lb-0-642-yolov8l-yolov11-ensemble) which has been well used in this competition. Also check out my colleague Jackson's [notebook](https://www.kaggle.com/code/jacksonpond/demo-notebook).", "3173436": "YOLO has an early stopping function in its arguments, so if training stops before reaching the set number of epochs, that is the cause.\n```\n# Number of epochs to wait without improvement in validation metrics before early stopping the training. \n# Helps prevent overfitting by stopping training when performance plateaus.\npatience=5, \n```", "3173531": "hideyukizushi I was trying to say that I am not seeing any improvement after 30-35 epochs, using adamW optimizer lr- 0.0005 \nFine tuning after 100 epochs gave 0.57 lb\nShould I perform some sort of preprocessing before training for improvement?", "3173721": "Yes, did you see this discussion? [Preprocessing approach by Nikolai Prudnikov](https://www.kaggle.com/competitions/byu-locating-bacterial-flagellar-motors-2025/discussion/569996)\nSeveral pre-processing techniques such as augmentation, denoising, normalization,etc can be tried out.", "3173728": "Thanks, I discovered the @jacksonpond Notebook" }, "source": "meta" }