Rethinking BiSeNet for Real-Time Semantic Segmentation

group convolution to reduce computation cost while main-taining comparable accuracy. These works are particularly designed for the image classification tasks, and their exten-sions to semantic segmentation application should be care-fully tuned. 2.2. Generic Semantic Segmentation Traditional segmentation algorithms, e.g., threshold se-

  Image, Rethinking, Segmentation, Convolutions, Semantics, Semantic segmentation