Hi, and thanks for sharing the inference code for RaCo.
If releasing the training code isn't possible, could you be a bit more specific about the augmentations used? The paper only states:
"Specifically, we generate training pairs using synthetic homographies with full 360° rotations combined with strong photometric transformations."
In particular, it would help to know the parameter ranges for the homographies (rotation, scale, perspective, translation) and which photometric transformations were applied along with their magnitudes.
Thanks!
Hi, and thanks for sharing the inference code for RaCo.
If releasing the training code isn't possible, could you be a bit more specific about the augmentations used? The paper only states:
"Specifically, we generate training pairs using synthetic homographies with full 360° rotations combined with strong photometric transformations."
In particular, it would help to know the parameter ranges for the homographies (rotation, scale, perspective, translation) and which photometric transformations were applied along with their magnitudes.
Thanks!