Georg Bökman

I have this issue as well. There is a check implemented that seems to relate to the problem: https://github.com/cvg/pixloc/blob/8f253dcd1f1d7cbe1bc4f1cb4ee628b1bd344fcb/pixloc/pixlib/geometry/optimization.py#L34-L45 However, the error message obtained when `cholesky` fails in torch=1.10 is...

I think this is because the LoFTR positional encoding has a max image size of 2048 pixels by default in a precomputed tensor for the encoding. I guess there are...

Yes, I can have a look.

@zehongs We have been able to reproduce the numbers in https://github.com/GrumpyZhou/image-matching-toolbox/issues/20 , if you could comment on/clarify the below points it would be very helpful: - Is it correct that...

Hi, I would also like to start by thanking you for making your great work available here. For me this issue is not present on the CPU, but only on...

> The signature for any TNN should pass ( features, neighborhoods) somehow. In pyg this has the format ( features, index). > > In tnx, we create a Data class...

Some more things about the training loop. 4. If as suggested above in 2., the `softmax` is removed, then the checks `y_hat > 0.5` need to be replaced by `y_hat...

Hi Shuai, Thanks for the eval! Here's a demo for the learned steerer: [colab](https://colab.research.google.com/drive/1sCqgi3yo3OuxA8VX_jPUt5ImHDmEajsZ?usp=sharing). I think one needs to tune the `min_cossim_coarse` threshold, perhaps. Combining with e2cnn would definitely also...

Hi @noahzn, I think there are different options, we discussed it a bit here: https://github.com/georg-bn/rotation-steerers/issues/3 If you are interested in working on this as a research project, let me know...