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
6.38k stars 490 forks source link

Does this work with xformers? #41

Open brentjohnston opened 5 months ago

brentjohnston commented 5 months ago

I am working on Unity project to test this model out for converting 2D images and video to 3D (it uses beit currently and works great). It was pretty slow with the 2D to 3D processing of video with this model though (like 2 fps)

so I tried to pip install xformers for a potential speedup (I was hoping), but it says this version of xformers I'm using is not compatible with Torch 2.0.1+cu117 (Cuda 11.7).

The only version of xformers that would work that I found would be version 22 and potentially working with only cuda 11.8 and torch 2.0.1. Is it okay to use cuda 11.8 with this model, I think the Unity project requires 11.7. Does anyone know an old xformers whl that works with Torch 2.0.1+cu117 Also would xformers even make a difference for the framerate? Thanks.

Are there any other parameters in the .py files I can adjust to increase the framerate? It seems like all 3 models run at 2 fps.

LiheYoung commented 5 months ago

Sorry, I do not have enough experience in installing xformers. Hope others can help :)

LiheYoung commented 5 months ago

Hi @brentjohnston, have you tried torch 2.0.0+cu117 and xformers 0.0.18?

antimattergroup commented 4 months ago

Hi @brentjohnston, have you tried torch 2.0.0+cu117 and xformers 0.0.18?

Is this a recommended environment configuration? It always have 'xFormers not available' report in my console. Now I'm using torch 2.2.0+cu121 and xformers 0.0.24

jianchaoci commented 1 month ago

Hi @brentjohnston, have you tried torch 2.0.0+cu117 and xformers 0.0.18?

Is this a recommended environment configuration? It always have 'xFormers not available' report in my console. Now I'm using torch 2.2.0+cu121 and xformers 0.0.24

Did you solve the problem, I am facing the same problem.