search

KaiqiangXiong / CL-MVSNet

[ICCV2023] CL-MVSNet: Unsupervised Multi-view Stereo with Dual-level Contrastive Learning
MIT License
40 stars 3 forks source link

About the performance of using unity as depth representation #5

Open averyhe opened 4 months ago

averyhe commented 4 months ago

Hi there, Thanks for your excellent work! I'm trying to use other representations of depth (such as the one used in UniMVSNet) and happened to notice that there's a code snippet of function 'unity_regression'. I'm interested in why this function is not used in this work and what's the performance of it? It would be greatly appreciated if you could share some observations on this topic. Thank you!

KaiqiangXiong commented 3 months ago

As a form of deep representation, Unification effectively combines the advantages of classification and regression, demonstrating superior performance in supervised MVS. However, there are some issues when it comes to unsupervised algorithms. Due to the integration of classification-based representation, gradients can only be propagated back in the region of offset, where the regression works, resulting in a weaker supervisory signal than normal regression representations.

averyhe commented 3 months ago

Got it. Thanks a lot for your quick and insightful reply!