Dmytro Mishkin

Situation is getting better with version `torch==1.13.0.dev20220521` and running with `PYTORCH_ENABLE_MPS_FALLBACK=1`. But still there are crashes

https://github.com/pytorch/pytorch/issues/78247

https://github.com/pytorch/pytorch/issues/85143 - this blocks a lot of kornia functionality, by crashing Other than that, which can be temporarily-fixed on our side: - [x] spatial gradient relies on conv3d, which is...

@edgarriba I will do the things in separate PRs once the basic for testing https://github.com/kornia/kornia/pull/1716 be merged

@edgarriba spatial gradient is fixed here https://github.com/kornia/kornia/pull/1898 There is also "spatial_gradient3d", which is not fixed, but there 3d convolution is actually correct alg and it is much rarery used

https://github.com/pytorch/pytorch/issues/86107

What we also can do, is to anisotropic mip-mapping (aka scale-pyramid) https://en.wikipedia.org/wiki/Anisotropic_filtering, specifically this https://en.wikipedia.org/wiki/Anisotropic_filtering#/media/File:MipMap_Example_STS101_Anisotropic.png Then we sample from proper locations. I guess it would be useful for many things....

@Skydes PR is not needed, I wanted confirmation :) https://github.com/kornia/kornia/pull/1616

@Skydes since the thing was fixed in kornia, how about removing code from hloc and rely on kornia instead? I can make a PR if needed.

Which OS are you running on?