Open lvgeng opened 3 years ago
Hi What do you mean with "60% overlapping" ? Also, why do you think a feature descriptor would have trouble dealing with small images? R2D2 computes patch descriptors with a receptive field of ~40x40 pixels, so as long as the image is larger than that, it's fine. See also #30
It is more like a general question... Considering the problem I am working on (light field), I wish to use R2D2 with low resolution images (like 40 by 40, and the images are supposed to have more than 60% overlapping), but it seems to be hard for a feature exporter to work in low resolution. Any suggestion on how to tweak?