search

TRI-ML / packnet-sfm

TRI-ML Monocular Depth Estimation Repository
https://tri-ml.github.io/packnet-sfm/
MIT License
1.23k stars 242 forks source link

Failure cases #160

Closed jsczzzk closed 3 years ago

jsczzzk commented 3 years ago

eigen_test_files_0000000019 eigen_test_files_0000000021 I found that the PackNet-SfM cannot accurately estimate the depth of the man in white. The depth of his upper body is consistent with the background. What is the reason? Thanks in advance.

VitorGuizilini-TRI commented 3 years ago

Self-supervised learning methods have issues with dynamic objects, that's why you are seeing those artifacts. There are some techniques that can mitigate that, and we use a few (automasking, minimum reprojection error), however it does not completely eliminate this issue.

jsczzzk commented 3 years ago

Thank you so much!