Does YOLOV5 classification support the `resume training`

[AI-Passionner]

Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

See above.
Thanks.

Additional

No response

[pourmand1376]

Yes, use --resume parameter when training. Also, you can load your previous trained models using --weight parameter and use it to resume your training.

[AI-Passionner]

@pourmand1376 Thank you.
That works for object detection. Classification might not support yet.

[pourmand1376]

Hmm. That I don't know! Thanks for further clarification. @AI-Passionner

[glenn-jocher]

TODO: Classification --resume

[RamatovInomjon]

TODO: Classification --resume

Hi , Thanks for the great work! Classification resume is done?

[glenn-jocher]

@RamatovInomjon no, not started yet (extremely overloaded with tasks). If you'd like to contribute an implementation PR that would really help.

The fastest and easiest way to incorporate your ideas into the official codebase is to submit a Pull Request (PR) implementing your idea, and if applicable providing before and after profiling/inference/training results to help us understand the improvement your feature provides. This allows us to directly see the changes in the code and to understand how they affect workflows and performance.

Please see our ✅ Contributing Guide to get started.

[IyazGreat]

HI @glenn-jocher , I trust you are well.

Any update on this?

Thanks!

[glenn-jocher]

@IyazGreat Hi there! Unfortunately, the classification resume feature has not been implemented yet. We are currently busy with various tasks, but we are always open to community contributions. If you'd like to help with this feature, please consider submitting a Pull Request following our ✅ Contributing Guide. Your input would be greatly appreciated.

[Blessing988]

I have trained my YOLOv5 classication model for some epochs and I want to resume it. The ----resume argument isn't in the training script same as the weights argument. How do I continue training the model?

[glenn-jocher]

@Blessing988 for resuming training of your YOLOv5 classification model, while the --resume argument is specifically designed for object detection tasks and might not be directly applicable for classification tasks in its current form, you can still effectively "resume" your training by using the --weights argument. Here's how you can do it:

1. When you initially start your training, YOLOv5 saves checkpoints of your model at regular intervals in the runs/train/exp*/weights directory (where * is the experiment number).
2. To resume training, you can use the last saved checkpoint by specifying it with the --weights parameter. For example, if your last checkpoint is best.pt, you would use it like this:

python train.py --img 640 --batch 16 --epochs 100 --data your_dataset.yaml --weights runs/train/exp*/weights/best.pt

This command tells YOLOv5 to start training using the weights from your last checkpoint, effectively resuming your training.

Remember, while this method allows you to continue training from where you left off, it's always a good idea to keep an eye on your training metrics to ensure everything is proceeding as expected.

[Blessing988]

@glenn-jocher Thank you very much.

[glenn-jocher]

@Blessing988 you're welcome! If you have any more questions or need further assistance as you continue with your project, feel free to reach out. Happy training! 🚀

[Blessing988]

@glenn-jocher Also for YOLOv5 classification model, does it have the same backbone and neck as the ones for object detection? If yes, how many conv blocks are in the head.

Can you let me know the architecture overview of the classification models as I can't find the yaml file for it. I would be glad. Thanks.