Detecting Out-of-distribution Examples via Class-conditional Impressions Reappearing
Jinggang Chen (Huazhong University of Science and Technology); Xiaoyang Qu (Ping An Technology (Shenzhen) Co., Ltd); Junjie Li (Huazhong University of Science and Technology); Jianzong Wang (Ping An Technology (Shenzhen) Co., Ltd); Jiguang Wan (Huazhong University of Science and Technology); Jing Xiao (Ping An Insurance (Group) Company of China)
-
SPS
IEEE Members: $11.00
Non-members: $15.00
Out-of-distribution (OOD) detection aims at enhancing standard deep neural networks to distinguish anomalous inputs from original training data. Previous progress has introduced various approaches where the in-distribution training data and even several OOD examples are prerequisites. However, due to privacy and security, auxiliary data tends to be impractical in a real-world scenario. In this paper, we propose a data-free method without training on natural data, called Class-Conditional Impressions Reappearing (C2IR), which utilizes image impressions from the fixed model to recover class-conditional feature statistics. Based on that, we introduce Integral Probability Metrics to estimate layer-wise class-conditional deviations and obtain layer weights by Measuring Gradient-based Importance (MGI). The experiments verify the effectiveness of our method and indicate that C2IR outperforms other post-hoc methods and reaches comparable performance to the full access (ID and OOD) detection method, especially in the far-OOD dataset (SVHN).