Hey, it seems that optica flow labels are incorrect: I tried running the default example with the torch Dataset, and I see optical flow that is composed into a grid, in which each cell of some H and W has only one unique value. So doing len(flow.unique()) will give about 50-200 unique values. Is something wrong in how the labels were written/read? Please provide some comments on that.
Note: I read other discussions on the optical flow labels in the dataset and I know about the value range, that's not the problem right now.
I opened: discrete/images/train/front
#52
Hey, it seems that optica flow labels are incorrect: I tried running the default example with the torch Dataset, and I see optical flow that is composed into a grid, in which each cell of some H and W has only one unique value. So doing
len(flow.unique())will give about 50-200 unique values. Is something wrong in how the labels were written/read? Please provide some comments on that.Note: I read other discussions on the optical flow labels in the dataset and I know about the value range, that's not the problem right now.
I opened: discrete/images/train/front
#52