Just to close the discussion with a bit of info.
i tried, torchvision.resize(), torch.nn.interpolate(), PIL.Image.resize(), cv2.resize(), and tensorflow.image.resize()
torchvision.resize() and torch.nn.interpolate() return image diff of 0.
torch 1.5 and tensorflow 2.2 have a slight differetnce(image attachhed)
torch 1.8 and tensorflow 2.2 are identical -
all others are different from each other! curiously, since models tend to learn surface, they will all return slightly different benchmarks. Which is what led me to look at this.