search

autonomousvision / unimatch

[TPAMI'23] Unifying Flow, Stereo and Depth Estimation
https://haofeixu.github.io/unimatch/
MIT License
1.05k stars 106 forks source link

Computing pose.txt #7

Closed Yijia-Xiao closed 1 year ago

Yijia-Xiao commented 1 year ago

Hi,

I am using this work to finetune on some depth images and make prediction.

However, I encountered some difficulty when trying to compute pose.txt for images. In your code, it seems that you are using colmap to get the pose.txt file for each image. I referred to colmap but couldn't figure out how to do that (issue in colmap). Could you provide some details/scripts on that?

BTW, I wonder if there will be an approach to directly predict depth from monocular images? It seems that current inference code needs consecutive images from one scene evaluate_depth.py#L338.

Thank you!

haofeixu commented 1 year ago

Besides the instruction you mentioned in colmap repo, you can also try this one: https://github.com/fyusion/llff#1-recover-camera-poses

Our method only supports two or more posed images as input for depth prediction. If you are looking for single image depth estimation methods, you might want to try https://github.com/isl-org/DPT

Yijia-Xiao commented 1 year ago

Hi @haofeixu, Thank you for your detailed reply:thumbsup:. I will check it out.

Yijia-Xiao commented 1 year ago

Hi @haofeixu ,

I tried https://github.com/fyusion/llff#1-recover-camera-poses, but got error message ERROR: failed to create sparse model. I am wondering if it would be convenient to share the colmap/llff scripts you used to extract the pose for https://github.com/autonomousvision/unimatch/tree/master/demo/depth-scannet (from color images and intrinsic to pose).

neverstoplearn commented 1 year ago

Hi @haofeixu ,

I tried https://github.com/fyusion/llff#1-recover-camera-poses, but got error message ERROR: failed to create sparse model. I am wondering if it would be convenient to share the colmap/llff scripts you used to extract the pose for https://github.com/autonomousvision/unimatch/tree/master/demo/depth-scannet (from color images and intrinsic to pose).

have you solve that,I meet save error.