Add return statement

[PatrickBue]

Adding simple 'return' statement to the train_one_epoch() function.

Fyi, we (Microsoft) are using the references/detection folder as-is in our Computer Vision Best Practices repository. All our Jupyter notebooks and higher-level functionality for Object Detection is based fully on torchvision.

See these notebooks:
https://github.com/microsoft/computervision-recipes/tree/master/scenarios/detection

And Torchvision's reference folder is copy & pasted here:
https://github.com/microsoft/computervision-recipes/tree/master/utils_cv/detection/references

[fmassa]

Thanks a lot!

Great to know that torchvision is being helpful for providing those reference implementations!

Would love to hear any feedback you might have!

[PatrickBue]

@fmassa Thanks for the quick sign-off!

More of a general question: a lot of the notebooks we implemented I believe would be very helpful for somebody who wants to use Torchvision for object detection. For example:

• https://github.com/microsoft/computervision-recipes/blob/master/scenarios/detection/01_training_introduction.ipynb
• https://github.com/microsoft/computervision-recipes/blob/master/scenarios/detection/04_coco_accuracy_vs_speed.ipynb

Currently, the references folder lacks a bit with regards to e.g. explanations or being beginner-friendly. Is there anywhere on Torchvision with links to resources such as our repo? Possible options could maybe be:

• To add a link+screenshot here (happy to start a PR) but not sure if that is aligned with torchvision's vision: https://github.com/pytorch/vision/tree/master/references/detection/README.md
• Or similarly add to: https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html