multi-gpu validation

[yjseok]

Search before asking

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

Question

Hello,

Is there any way to use multi-gpu during running val.py of Yolov5?
I set --device parameter as 0,1,2,3, but it doesn't work

Additional

No response

[glenn-jocher]

@yjseok hello,

Thank you for reaching out and for your thorough search before posting your question! Currently, val.py in YOLOv5 does not support multi-GPU validation directly through the --device parameter. The multi-GPU functionality is primarily designed for training purposes using train.py.

However, you can achieve multi-GPU validation by modifying the code to use torch.nn.DataParallel or torch.nn.parallel.DistributedDataParallel. Here’s a brief guide on how you might approach this:

1. Clone the YOLOv5 repository and install dependencies:

   git clone https://github.com/ultralytics/yolov5
   cd yolov5
   pip install -r requirements.txt

2. Modify val.py to use torch.nn.DataParallel:

   import torch
   from models.yolo import Model
   
   # Load model
   model = Model(cfg='yolov5s.yaml', ch=3, nc=80).to(device)
   
   # Wrap model with DataParallel
   if torch.cuda.device_count() > 1:
       model = torch.nn.DataParallel(model, device_ids=[0, 1, 2, 3])
   
   # Validate
   results, maps, times = validate(model=model, ...)

3. Run the modified val.py:

   python val.py --data coco.yaml --weights yolov5s.pt --device 0,1,2,3

For more detailed instructions and advanced configurations, you can refer to our Multi-GPU Training Guide.

If you encounter any issues or have further questions, please provide a minimum reproducible example of your code and the specific error messages you are seeing. This will help us to better understand and address your issue. You can find more information on creating a minimum reproducible example here.

Lastly, please ensure you are using the latest versions of torch and the YOLOv5 repository to avoid any compatibility issues.