bytedance / particle-sfm

ParticleSfM: Exploiting Dense Point Trajectories for Localizing Moving Cameras in the Wild. ECCV 2022.
http://b1ueber2y.me/projects/ParticleSfM/
GNU General Public License v3.0
277 stars 23 forks source link

Replace Midas with more recent depth models e.g., ZoeDepth, Metric3d-v2, marigold, etc. #26

Closed umariqb closed 6 months ago

umariqb commented 6 months ago

Hi,

Thank you so much for the great work. I wonder if replacing MIDAS with more recent models will help improve the SfM quality. Any thoughts?

Thanks!

guohengkai commented 6 months ago

Hi,

Applying more recent depth models will improve the motion segmentation (then more correct points for SfM) because we only use the depth estimation in the motion segmentation module. But it is possible to use the depth to help the initialization and optimization for SfM, like https://arxiv.org/abs/2204.02509.

umariqb commented 6 months ago

Thank you for the explanation and reference!