Is prediction speed of YOLOv5 recorded somewhere as part of the integration?

iankelk · May 14, 2022, 1:03am

Hello!
I’ve been training YOLOv5 models using wandb, and I’ve been amazed at how much is natively built in with just a few command-line arguments. One thing I’m lacking - is there any recording of the speed at which it makes predictions? I don’t particularly care if it’s on the training or test sets, I’m just looking for any record of the speed at which the model makes predictions.

Thanks!
Ian

_scott · May 14, 2022, 8:01am

Hi,

This is probably a YoloV5 specific question, but in general, in cases like these where you want an additional logged metric, you would just record it yourself and call wandb.log({'speed': speed}, commit=false).
That being said, I had a look at the yolov5 docs and it looks like you can pass —profile to get it to profile the inference speed of your model. Hope this helps

iankelk · May 14, 2022, 8:17pm

Hi Scott thanks for your reply! I’ll try using wandb.log({'speed': speed}, commit=false)

I looked in the docs here and didn’t see any reference to a -profile option, nor is it in the source code for train.py for YOLOv5. Where did you see it mentioned? Is it a different script to run other than train.py?

iankelk · May 14, 2022, 8:25pm

It seems that speed is not an option? I tried it and received an error.

----> 6 wandb.log({‘speed’: speed}, commit=false)

NameError: name ‘speed’ is not defined

_scott · May 14, 2022, 9:22pm

I’ll reach out to the author of YoloV5 and find the appropriate command to profile the inference speed.

_scott · May 14, 2022, 9:25pm

In the meantime, this tutorial profiles some different configurations.
https://colab.research.google.com/github/ultralytics/yolov5/blob/master/tutorial.ipynb#scrollTo=zR9ZbuQCH7FX

_scott · May 20, 2022, 12:09pm

It looks like you can call yolo.py directly with --profile, but not train.py.

github.com

ultralytics/yolov5/blob/fb7fa5be8b98b1f5f8162f699a10b2ddde5143c0/models/yolo.py#L311

      
        
                        ch = []
                    ch.append(c2)
                return nn.Sequential(*layers), sorted(save)
            
            

            
if __name__ == '__main__':
                parser = argparse.ArgumentParser()
                parser.add_argument('--cfg', type=str, default='yolov5s.yaml', help='model.yaml')
                parser.add_argument('--batch-size', type=int, default=1, help='total batch size for all GPUs')
                parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
                parser.add_argument('--profile', action='store_true', help='profile model speed')
                parser.add_argument('--line-profile', action='store_true', help='profile model speed layer by layer')
                parser.add_argument('--test', action='store_true', help='test all yolo*.yaml')
                opt = parser.parse_args()
                opt.cfg = check_yaml(opt.cfg)  # check YAML
                print_args(vars(opt))
                device = select_device(opt.device)
            
            
    # Create model
                im = torch.rand(opt.batch_size, 3, 640, 640).to(device)
                model = Model(opt.cfg).to(device)

lesliewandb · May 26, 2022, 1:13pm

Hi Ian, since we have not heard back from you we are going to close this request. If you would like to re-open the conversation, please let us know!

system · July 19, 2022, 12:10pm

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.

Topic		Replies	Views
Easy integration with yolov8? W&B Help	4	573	March 14, 2023
Horrible performance when viewing charts for WandB run W&B Help dashboard , wandb , pytorch	4	477	April 6, 2023
Wandb.watch not logging parameters W&B Help	19	1342	February 5, 2022
How do I do probabilistic logging? W&B Help	3	209	September 20, 2022
How can I watch layer activations? W&B Help	8	758	March 8, 2024

Is prediction speed of YOLOv5 recorded somewhere as part of the integration?

Related Topics