FBNet: Feature Balance Network for Urban-Scene Segmentation

Lei Gan, Huabin Huang, Banghuai Li, Ye Yuan

Abstract: Image segmentation in the urban scene has recently attracted much attention due to its success in autonomous driving sys-tems. However, the poor performance of concerned fore-ground targets, e.g., traffic lights and poles, still limits its further practical applications. In urban scenes, foreground targets are always concealed in their surrounding stuff be-cause of the special camera position and 3D perspective pro-jection. What's worse, it exacerbates the imbalance between foreground and background classes in high-level features due to the continuous expansion of the receptive field. We call it Feature Camouflage. In this paper, we present a novel add-on module, named Feature Balance Network (FBNet), to eliminate the feature camouflage in urban-scene segmen-tation. FBNet consists of two key components, i.e., Block-wise BCE(BwBCE) and Dual Feature Modulator(DFM). Bw-BCE serves as an auxiliary loss to ensure uniform gradients for foreground classes and their surroundings during back-propagation. At the same time, DFM intends to enhance the deep representation of foreground classes in high-level features adaptively under the supervision of BwBCE. These two modules facilitate each other as a whole to ease feature cam-ouflage effectively. Our proposed method achieves a new state-of-the-art segmentation performance on two challenging urban-scene benchmarks, i.e., Cityscapes and BDD100K. Code will be released for reproduction.