search

LiheYoung / Depth-Anything

[CVPR 2024] Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data. Foundation Model for Monocular Depth Estimation
https://depth-anything.github.io
Apache License 2.0
6.82k stars 523 forks source link

Converting disparities to depths #75

Open maturk opened 7 months ago

maturk commented 7 months ago

Hi

I am experimenting with using DepthAnything for depth supervision (not inverse depth == disparity). I am wondering if you have any advice in converting disparity estimates to relative depths, because I have some doubts. I tried something simple like depth = 1 / output_disparity.clamp(min=0.01) but I have some doubts because I get quite different looking depth maps compared to Zoe Depth on some of the indoor datasets I am playing with:

Inverse of DepthAnything: image

Zoe depth output: image

Let me know if you have any recommendations. Thanks.

cognitiveplus commented 7 months ago

Hi, have you tried applying gamma correction to DA output? like np.power (value, 1.0 / 2.2) You might also find this issue relevant: https://github.com/LiheYoung/Depth-Anything/issues/47