Dmytro Mishkin
Dmytro Mishkin
The workaround I am using now, is to start from random weights, then load a one of DPT-for depth from Facebook and initialize the backbone from there. ```python3 model_nyu =...
Well, for me, the "feature" is "works on my M1 machine with minimal changes". I don't have any Intel clang, unfortunately. Let me check if that works w/o FindPythonPyEnv and...
@Skydes I have updated PR
Hi, Unfortunately, RANSAC is a stochastic algorithm by definition. What you could do is to increase the number of iterations (max_iter)
You may want to use `pv.start_xvfb()` to create a virtual display
`H, inliers = cv2.findHomography(m_kpts0, m_kpts1, cv2.USAC_MAGSAC, 0.5, 0.999, 100000)`
Yes, we are going to add PoseLib RANSAC to our homography benchmark. https://github.com/danini/homography-benchmark/tree/main We actually added it code-wise, but have to tune threshold to get meaningful results.
@Skydes @Phil26AT Thank you! I will update the PR and merge into kornia tomorrow or Thursday
The output of AffNet is the "canonical shape" of the local feature (think of SIFT), not the transformation between images. Please, check the AffNet paper (or any other local feature...
https://kornia-tutorials.readthedocs.io/en/latest/_nbs/image_matching_adalam.html Check this image from AffNet example. The ellipses in each image are what predicted in AffNet. Then we match them based on local descriptor, thus establishing the correspondence. 