Open catid opened 9 months ago
Are you running this on GPU? I'm unsure whether parallelizing calls on the same GPU is safe. If all images have the same size (which is most often the case), we could instead batch the inference - this is safer and more efficient.
Sounds good
This commit works: https://github.com/catid/pixel-perfect-sfm/commit/89f487fbb0cb9d08cfe8ed16f8deda39501ba049