mAP zero while training

[rexayyy]

Search before asking

• I have searched the HUB issues and found no similar bug report.

HUB Component

Training

Bug

@glenn-jocher the issue still prevails as i have tried training using a different dataset and even different model and tried both hub and custom colab notebook

Environment

No response

Minimal Reproducible Example

No response

Additional

No response

[glenn-jocher]

@rexayyy this isn't a HUB issue, this is related to YOLOv5 and has been resolved in ultralytics/yolov5#9037, so all you need to do is start from a fresh Colab notebook or update your local YOLOv5, i.e. git pull from inside your YOLOv5 directory.

[rexayyy]

I will try now and let you know right now

[rexayyy]

it is still the same

[glenn-jocher]

@rexayyy yes I'm able to reproduce. Am investigating now.

[glenn-jocher]

@rexayyy I was able to fix a zero-map issue in ultralytics/yolov5#9056, but after training a new HUB model I'm getting a strange effect of later epochs showing zero map but the final val run showing correct mAP. It seems mAP-during training is still incorrect as in my original screenshot.

To receive this update:

• Git – git pull from within your yolov5/ directory or git clone https://github.com/ultralytics/yolov5 again
• PyTorch Hub – Force-reload model = torch.hub.load('ultralytics/yolov5', 'yolov5s', force_reload=True)
• Notebooks – View updated notebooks
• Docker – sudo docker pull ultralytics/yolov5:latest to update your image

Thank you for spotting this issue and informing us of the problem. Please let us know if this update resolves the issue for you, and feel free to inform us of any other issues you discover or feature requests that come to mind. Happy trainings with YOLOv5 🚀!

[rexayyy]

I will do it again and let you know the progress

[rexayyy]

I am still getting the issue

[MGIbrahim]

I am also having the same problem now

[glenn-jocher]

@rexayyy @MGIbrahim sorry about the persistent problems. I think I've resolved this now in ultralytics/yolov5#9058

I created a new HUB model and trained in a new Colab notebook (YOLOv5n on COCO128 for 10 epochs) and everything looks fine now.

[MGIbrahim]

@glenn-jocher Thanks for your quick response on this! I did a git pull and tried training again, but it still shows the mAP of 0.

[rexayyy]

Yes, it is still not fixed. I don`t know what has happened as it was working totally fine a few hours ago.

[glenn-jocher]

@MGIbrahim @rexayyy zero-mAP problem is being tracked down in ultralytics/yolov5#9059

[robotaiguy]

@glenn-jocher do you notice that in the hub display in the above post, his Images and Labels coutns are messed up like mine are? But in the Hub image a few posts up, they're normal. And I haven't used Hub yet, only Colab and local. I wonder if there's some off-by-1 error. But it shouldn't be with Hub, because I have yet to use it. I'm looking forward to trying it, but I haven't yet.