OpenCIL: Benchmarking out-of-distribution detection in class incremental learning

Class incremental learning (CIL) aims to learn a model that can not only incrementally accommodate new classes, but also maintain the learned knowledge of old classes. Out-of-distribution (OOD) detection in CIL is to retain this incremental learning ability, while being able to reject unknown samples that are drawn from different distributions of the learned classes. This capability is crucial to the safety of deploying CIL models in open worlds. However, despite remarkable advancements in the respective CIL and OOD detection, there lacks a systematic and large-scale benchmark to assess the capability of advanced CIL models in detecting OOD samples. To fill this gap, in this study we design a comprehensive empirical study to establish such a benchmark, named OpenCIL, offering a unified protocol for enabling CIL models with different OOD detectors using two principled OOD detection frameworks. One key observation we find through our comprehensive evaluation is that the CIL models can be severely biased towards the OOD samples and newly added classes when they are exposed to open environments. Motivated by this, we further propose a novel approach for OOD detection in CIL, namely Bi-directional Energy Regularization (BER), which is specially designed to mitigate these two biases in different CIL models by having energy regularization on both old and new classes. Extensive experiments show that BER can substantially improve the OOD detection capability across a range of CIL models, achieving state-of-the-art performance on the OpenCIL benchmark. All codes and datasets are open-source at https://github.com/mala-lab/OpenCIL.