Area calculation in TorchVision Object Detection Finetuning Tutorial

[Himanshunitrr]

I see that at https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html,

area = (boxes[:, 3] - boxes[:, 1]) * (boxes[:, 2] - boxes[:, 0]) but shouldn't it be something like area = ((boxes[:, 3] - boxes[:, 1]) + 1) * ((boxes[:, 2] - boxes[:, 0]) + 1) for calculating areas? If I am wrong, can someone explain me why is it so. Thanks in advance !

cc @datumbox @nairbv

[NicolasHug]

Thanks for the report @Himanshunitrr

In torchvision, we compute the area without the +1 offset:

https://github.com/pytorch/vision/blob/5dd95944c609ac399743fa843ddb7b83780512b3/torchvision/ops/boxes.py#L219-L235

You can find more context in this thread: pytorch/vision#826 (comment)

I don't think we'll change the computation to add the +1 now, so I think we can close this issue (and the PR)

CC @svekars I don't have the permission to close :)