Can't get custom dataset to perform as well as monodepth2

[d4l3k]

I'm running into issues getting manydepth to produce good results with a custom dataset. The same dataset on monodepth2 works pretty well (though dynamic objects are incorrect). I'm using the same camera intrinsics as for monodepth2 so pretty sure they're correct. I've tried freezing the teacher it at 5 epochs as well but it produces the same results.

I've got about 26k pairs of images at 20 fps at 640x416 resolution.

python -m manydepth.train --dataset <custom> --data_path ../my-dataset/ --batch_size 9 --log_frequency 5 --height 416 --num_workers 16 --png --freeze_teacher_epoch 5

Original:

Manydepth:

Monodepth2:

[d4l3k]

Here's a slightly different dataset w/ the cost vol. With some model tweaks it's crisper but still not great. It seems to have the same issue with dynamic objects appearing as infinite as monodepth2.

[daniyar-niantic]

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?