search

patriciogonzalezvivo / prisma

Computational photography pipeline that performs multiple inferences from any image or video.
Other
204 stars 8 forks source link

DepthAnything for depth map building #2

Open viktordanov opened 7 months ago

viktordanov commented 7 months ago

I wanted to bring the following recent paper to this repo's maintainers. It is released by TikTok et al. which could be a relevant inclusion in the pipeline to support better realtime processing: https://github.com/LiheYoung/Depth-Anything. Upon superfluous review and research, it appears the new approach presented in the Depth Anything paper significantly outperforms MiDaS v3.1 and is worth consideration.

HuggingFace playground: https://huggingface.co/spaces/LiheYoung/Depth-Anything?ref=maginative.com

GitHub page of the project: https://depth-anything.github.io/?ref=maginative.com

patriciogonzalezvivo commented 7 months ago

Thank you @viktordanov ! I just added it https://github.com/patriciogonzalezvivo/prisma/blob/main/bands/depth_anything.py

patriciogonzalezvivo commented 7 months ago

Realize that what I added is only the relative depth, need to see how to use the metric. Maybe @LiheYoung could share a tip or two on how to do that?