Convert tensor to image in pytorch

image_tensor is a 3D tensor with devices: cudaimage_numpy = image_tensor[0].cpu().float().numpy() # convert it into a numpy arrayimage_numpy = (np.transpose(image_numpy, (1, 2, 0)) + 1) / 2.0 * 255.0 # post-processing: tranpose and scalingimage_pil = Image.fromarray(image_numpy)image_pil.save(image_path)