autonomousvision / unimatch

[TPAMI'23] Unifying Flow, Stereo and Depth Estimation
https://haofeixu.github.io/unimatch/
MIT License
965 stars 99 forks source link

FEATURE REQUEST: Support 'depth' task using two images with different Intrinsics. #30

Open JonathonLuiten opened 11 months ago

JonathonLuiten commented 11 months ago

I have images from different cameras with different intrinsics that I would like the calculate the depth from. I already have the intrinsics for each and the relative pose between them, but this code base doesn't support this currently (e.g. see matching.py line 260 and 265 which explicitly assumes the intrinsics are the same for both cameras.

Cheers, Jonathon

haofeixu commented 11 months ago

Thanks for your suggestion and you are welcome to send a PR to this repo, thanks.

Haobo-Liu commented 8 months ago

I have images from different cameras with different intrinsics that I would like the calculate the depth from. I already have the intrinsics for each and the relative pose between them, but this code base doesn't support this currently (e.g. see matching.py line 260 and 265 which explicitly assumes the intrinsics are the same for both cameras.

Cheers, Jonathon

Have you solved it?

JonathonLuiten commented 8 months ago

No sorry. In the end I just used a different code base for this.

Haobo-Liu commented 8 months ago

No sorry. In the end I just used a different code base for this.

Could you share the project you used?

JonathonLuiten commented 8 months ago

So I’m not 100% sure because I didn’t end up actually using this for any final results. I think I resorted to mono depth with Zoe Depth as an initial guess and then optimised to make it consistent between the frames with a nerf like variant.

This is VERY different to what your repo is doing, but it gave me approximately what I was looking for (depth consistent between two frames). If you ever implement this in your code would be cool.