Representation Compensation Networks for Continual Semantic Segmentation
This addresses the problem of incremental learning in semantic segmentation for computer vision applications, offering an incremental improvement over existing methods.
The paper tackles continual semantic segmentation by introducing a representation compensation module and pooled cube knowledge distillation to mitigate catastrophic forgetting when adding new classes, achieving state-of-the-art performance on two scenarios without extra inference costs.
In this work, we study the continual semantic segmentation problem, where the deep neural networks are required to incorporate new classes continually without catastrophic forgetting. We propose to use a structural re-parameterization mechanism, named representation compensation (RC) module, to decouple the representation learning of both old and new knowledge. The RC module consists of two dynamically evolved branches with one frozen and one trainable. Besides, we design a pooled cube knowledge distillation strategy on both spatial and channel dimensions to further enhance the plasticity and stability of the model. We conduct experiments on two challenging continual semantic segmentation scenarios, continual class segmentation and continual domain segmentation. Without any extra computational overhead and parameters during inference, our method outperforms state-of-the-art performance. The code is available at \url{https://github.com/zhangchbin/RCIL}.