`box_iou` doesn't work with degenerated boxes

🐛 Bug

torchvision.ops.boxes.box_iou works only if (x1,y1) < (x2,y2) and calculates zero IoU otherwise

To Reproduce

Steps to reproduce the behavior:

1. Consider a hypothetical bbox array:

hboxes = torch.arange(5,5*25,5).view(-1,4).float(); hboxes

tensor([[  5.,  10.,  15.,  20.],
        [ 25.,  30.,  35.,  40.],
        [ 45.,  50.,  55.,  60.],
        [ 65.,  70.,  75.,  80.],
        [ 85.,  90.,  95., 100.],
        [105., 110., 115., 120.]])

As all the (x1,y1) are less than (x2,y2) in this case, the box_iou output with itself is a perfect identity matrix

from torchvision.ops.boxes import box_iou

box_iou(hboxes,hboxes)

tensor([[1., 0., 0., 0., 0., 0.],
        [0., 1., 0., 0., 0., 0.],
        [0., 0., 1., 0., 0., 0.],
        [0., 0., 0., 1., 0., 0.],
        [0., 0., 0., 0., 1., 0.],
        [0., 0., 0., 0., 0., 1.]])

2. However, if the said condition doesn't hold (x1y1 > x2y2)

hboxes[[2,4],:] = torch.cat([hboxes[[2,4],2:],hboxes[[2,4],:2]],dim=1); hboxes

tensor([[  5.,  10.,  15.,  20.],
        [ 25.,  30.,  35.,  40.],
        [ 55.,  60.,  45.,  50.],
        [ 65.,  70.,  75.,  80.],
        [ 95., 100.,  85.,  90.],
        [105., 110., 115., 120.]])

like the row 2nd and 4th in this case

The box_iou output is 0

box_iou(hboxes,hboxes)

tensor([[1., 0., 0., 0., 0., 0.],
        [0., 1., 0., 0., 0., 0.],
        [0., 0., 0., 0., 0., 0.],
        [0., 0., 0., 1., 0., 0.],
        [0., 0., 0., 0., 0., 0.],
        [0., 0., 0., 0., 0., 1.]])

Expected behavior

The coordinates should be handled beforehand and output for the 2nd case mentioned above should also be an Identity Matrix

Environment

PyTorch version: 1.6.0
Is debug build: No
CUDA used to build PyTorch: 10.2

OS: Manjaro Linux
GCC version: (GCC) 10.1.0
Clang version: Could not collect
CMake version: version 3.17.3

Python version: 3.8 (64-bit runtime)
Is CUDA available: Yes
CUDA runtime version: Could not collect
GPU models and configuration: GPU 0: GeForce 940MX
Nvidia driver version: 440.100
cuDNN version: Could not collect

Versions of relevant libraries:
[pip3] numpy==1.19.1
[pip3] pytorch-lightning==0.7.6
[pip3] torch==1.6.0
[pip3] torchvision==0.7.0
[conda] blas 1.0 mkl
[conda] cudatoolkit 10.2.89 hfd86e86_1
[conda] mkl 2020.1 217
[conda] mkl-service 2.3.0 py38he904b0f_0
[conda] mkl_fft 1.1.0 py38h23d657b_0
[conda] mkl_random 1.1.1 py38h0573a6f_0
[conda] numpy 1.19.1 py38hbc911f0_0
[conda] numpy-base 1.19.1 py38hfa32c7d_0
[conda] pytorch 1.6.0 py3.8_cuda10.2.89_cudnn7.6.5_0 pytorch
[conda] pytorch-lightning 0.7.6 pypi_0 pypi
[conda] torchvision 0.7.0 py38_cu102 pytorch

Willing to work

I'd be glad to fix this behaviour 🙂

Start in torchvision/ops/boxes.py at box_iou and reproduce the provided Python examples with valid and degenerate boxes. Determine how degenerate coordinates should be handled, then add coverage for both cases and verify that self-comparison produces the expected identity values.