Densepose DensePoseOutputsTextureVisualizer.visualize fails with torch==2.1.0

[pjbull]

Instructions To Reproduce the 🐛 Bug:

With torch=2.1.0 DensePoseOutputsTextureVisualizer.visualize fails with type mismatch problem.

1. Full runnable code or full changes you made:

DensePoseOutputsTextureVisualizer().visualize(image_arr, outputs_boxes_xywh_classes)

Result:

TypeError: expected size to be one of int or Tuple[int] or Tuple[int, int] or Tuple[int, int, int], but got size with types [<class 'numpy.int32'>, <class 'numpy.int32'>]

Expected behavior:

No error.

Environment:

This seems to only happen with the latest torch=2.1.0

Cause:

This seems to be caused by the .numpy() call here: https://github.com/facebookresearch/detectron2/blob/main/projects/DensePose/densepose/vis/densepose_outputs_vertex.py#L193

The other visualizer does this with .int() instead, which should fix it: https://github.com/facebookresearch/detectron2/blob/main/projects/DensePose/densepose/vis/densepose_outputs_vertex.py#L80

[dsantiago]

+1

[Zakonildo]

+2