Daniyar Turmukhambetov

Hi @ysuwl Your question is lacking details, are you talking about mono-trained model? stereo-trained model? KITTI lidar depths are not used during training. There is median scaling protocol for mono-trained...

https://github.com/nianticlabs/panoptic-forecasting is available now

Hi @kjyothiswaroop ! The computation of surface normals was a custom code. We are working on releasing the code soon.

Hi @marichka-dobko The dataset is available now. The evaluation code will hopefully be available in 3 weeks.

Hi. Have you changed intrinsics?

Hi. Do you use single images for prediction, or multiple images?

With single image prediction this is probably acceptable quality of depth maps. You have different intrinsics then at training time, and there are no additional images to resolve depth ambiguities.

Oh, sorry I thought this is trained on KITTI. You are right, the intrinsics would be the same. I would suggest to make a multi-frame prediction and see if local/global...

Hi! I suspect you need to play with `freeze_teacher_epoch`. Why do you say you have 26k *pairs* of images, are you training on video or stereo?

Hi. Have you verified that the colmap reconstruction looks correct?