For the coordinate inputs for something like torchvision.ops.nms are the coordinates supposed to be in pixel or normalized coordinates? I know that the Tensorflow version of nms is normalized, but it isn’t explicitly stated in the Pytorch documentation.