Loss implementation seems different than equation (4) in the paper.

nv-tlabs / GSCNN

Gated-Shape CNN for Semantic Segmentation (ICCV 2019)

https://nv-tlabs.github.io/GSCNN/

Other

916 stars 200 forks source link

Loss implementation seems different than equation (4) in the paper. #44

Closed SHMCU closed 4 years ago

SHMCU commented 4 years ago

Looks like the loss implementation does not apply argmax to the semantic segmentation logits, right? But in the paper, equation (4) applies argmax before ConvTri and Gradient computation, right? Which one is the one used to produce the experimental results?

Thank you, Hai

SHMCU commented 4 years ago

The magnitude computation in custom_functions.py, the mag could become 1.0 for all black classes (no object of this class appear), right? Is this a bug? or intended?

tovacinni commented 4 years ago

Addressed in #46