`np.array(vid_feats)` saves the list into float64 which doubles the disk space. casting it to f32 (or, ideally, to a user-defined precision) would be nice. it is should be very simply to do - [ ] implement casting to f32 - [ ] (optional) make it configurable with f32 as default
np.array(vid_feats)saves the list into float64 which doubles the disk space.casting it to f32 (or, ideally, to a user-defined precision) would be nice.
it is should be very simply to do