Retrieving trained YOLOv5 weights from WandB

[aniruddh10124]

Search before asking

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

Question

After training the weights on YOLOv5 and saving the results on WandB, how do I save and retrieve those weights from WandB for using the trained weights to detect images in the test set? I know that some change needs to be done in the argument of !python detect.py --weight <argument>, but what should I write in place of <argument>?
Just to be clear, I am able to retrieve the trained weights from Google Drive, but my teams requires the weights to be stored and retrieved from WandB to streamline our work, which I am not able to do.
Thank you for your time.

Additional

No response

[github-actions]

👋 Hello @aniruddh10124, thank you for your interest in YOLOv5 🚀! Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution.

If this is a 🐛 Bug Report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset images, training logs, screenshots, and a public link to online W&B logging if available.

For business inquiries or professional support requests please visit https://ultralytics.com or email Glenn Jocher at [email protected].

Requirements

Python>=3.6.0 with all requirements.txt installed including PyTorch>=1.7. To get started:

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

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Google Colab and Kaggle notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training (train.py), validation (val.py), inference (detect.py) and export (export.py) on MacOS, Windows, and Ubuntu every 24 hours and on every commit.

[glenn-jocher]

@aniruddh10124 thanks for your question! Our W&B expert @AyushExel should be able to help here.

[AyushExel]

@aniruddh10124 Hey, are you asking if you can directly pass artifact links to perform inference like you can to resume runs?
python detect.py --weights wandb-artifact://{artifact link} ? This isn't supported currently. wandb is only supported in training tasks.

How can you access the weights uploaded as W&B artifact?

Each artifact comes with an API. If you go to your dashboard where model is saved, you should see something like this:

You can copy and execute this api in your system and it'll download the weights. You can then use the the weights normally. Example:
python detect.py --weight {path to downloaded model}

Hope this helps :)

[aniruddh10124]

@AyushExel So, I guess what we are trying to do is not currently supported by WandB. Thank you for your clarification.

[AyushExel]

@glenn-jocher are there any plans to support passing weights links for inference as Anirudh is describing? Not just for artifacts but also things like s3 bucket, GCP etc. links

[glenn-jocher]

@AyushExel that's funny, that's probably the only argument that doesn't allow direct URL passing, because we already have the autodownload handling for the release assets. I tested it out just now and it seems to already be partially enabled, but the workflow has issues. I can take a look at it this week, I'll add a TODO.