Roman Shapovalov

I had similar problems. After downgrading Pytorch, it was not able to find the shared library. I then installed from scratch with Pytorch 1.4 in a new environment, and it...

Unfortunately we only crowd-sourced RGB videos without any depth signal or even knowledge of sensor calibration, so we cannot provide metric depth. The coordinate system was normalised for each scene...

Hi Prune, thanks for your interest in the dataset! For training vanilla NeRF you should not need depth maps – or are you trying some variant that uses pre-computed geometry?...

Thanks for the interest @angshine ! Before I can share the data, I have to anonymise them (blur license plates); I am trying to find time to do it.

Hello @Subin-Jeon , thanks for the interest and sorry for the delay. I am now talking to our legal team to clarify what derived data we can share given CelebA...

The scores are indeed the higher the better. They were both regressed based on COLMAP outputs, so probably do correlate.

Thanks for your interest. CO3Dv2 was augmented with more sequences, and there are now indeed 36k sequences in total.

Hi Andy, I’ve committed few follow-up edits. They concern double energies for varying edge potentials. The condition `if e.shape[0] == 3` in `cut_from_graph()` seemed buggy, I’ve changed to `shape[1]`. Roman

Andy, 2 more follow-up edits. 1. Fixed memory leak, which was crucial when functions were called repeatedly, e.g. during learning. 2. Added energy value returned by the library to the...

I tested it with your implementations of structured SVM — both cutting-plane and subgradient (thanks, BTW!). I switched from Joachims' SVM^struct code, because it was messy, used the QP solver...