search

lorenmt / mtan

The implementation of "End-to-End Multi-Task Learning with Attention" [CVPR 2019].
https://shikun.io/projects/multi-task-attention-network
MIT License
665 stars 108 forks source link

Did you mask out pixels where the ground truth is missing? #13

Closed ghost closed 4 years ago

ghost commented 4 years ago

Hello. Did you mask out pixels where the ground truth is missing for surface normal estimation and depth prediction as <Predicting Depth, Surface Normals and Semantic Labels with a Common Multi-scale Convolutional Architecture>?

lorenmt commented 4 years ago

Yes. Binary masks are applied to zero out the invalid gradients. Please check the loss functions implementation for more details.

On Mon, Oct 7, 2019 at 8:33, KejieLyu notifications@github.com wrote: Hello. Did you mask out pixels where the ground truth is missing for surface normal estimation and depth prediction as <Predicting Depth, Surface Normals and Semantic Labels with a Common Multi-scale Convolutional Architecture>?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub [https://github.com/lorenmt/mtan/issues/13?email_source=notifications&email_token=ABP2GHYAKXVXB65RWVAQL73QNJ7VVA5CNFSM4I56NKXKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HP57MHA] , or mute the thread [https://github.com/notifications/unsubscribe-auth/ABP2GH2M76UH6QMH73W7RU3QNJ7VVANCNFSM4I56NKXA] .

ghost commented 4 years ago

Thank you.