search

KexianHust / Structure-Guided-Ranking-Loss

Structure-Guided Ranking Loss for Single Image Depth Prediction
Other
184 stars 18 forks source link

Edge-Guided Sampling #10

Closed NekonekoHu closed 3 years ago

NekonekoHu commented 3 years ago

thanks for your great work! how did you achieve Edge-Guided Sampling? what does valid-mask mean?

KexianHust commented 3 years ago

For each edge point, we sample 4 points according to the Eqn.5 in the paper. Also, you can find the sampling details in the algorithm 1. The valid masks indicate the pixels whose GT depth values are valid. For those invalid GT, we should ignore their losses during training.