Optimized way to extract patches based on bounding box in PyTorch

Question

Assuming we got tensors: img and bounding boxes: bbox of shape:

[B, C, H, W], [B, xmin, ymin, xmax, ymax], respectively