Did you get the superpoint.npy in cache by using Superpoint-officially-pretrained model to extract features on the original image resolution, not 240320 or 480640 resolution? Or you retrain the Superpoint model ?
We use the official pretrained model on full resolution similar to other works (e.g., SuperGlue). All other parameters are the same as in the official repository.
Did you get the superpoint.npy in cache by using Superpoint-officially-pretrained model to extract features on the original image resolution, not 240320 or 480640 resolution? Or you retrain the Superpoint model ?