Knowledge Distillation from A Stronger Teacher
This work addresses a bottleneck in knowledge distillation for practitioners using advanced models, offering a simple and effective method to improve student model training from stronger teachers.
The paper tackles the problem of knowledge distillation from strong teacher models, where existing methods fail due to severe prediction discrepancies, and proposes a correlation-based loss that preserves inter-class and intra-class relations, achieving state-of-the-art performance across image classification, object detection, and semantic segmentation tasks.
Unlike existing knowledge distillation methods focus on the baseline settings, where the teacher models and training strategies are not that strong and competing as state-of-the-art approaches, this paper presents a method dubbed DIST to distill better from a stronger teacher. We empirically find that the discrepancy of predictions between the student and a stronger teacher may tend to be fairly severer. As a result, the exact match of predictions in KL divergence would disturb the training and make existing methods perform poorly. In this paper, we show that simply preserving the relations between the predictions of teacher and student would suffice, and propose a correlation-based loss to capture the intrinsic inter-class relations from the teacher explicitly. Besides, considering that different instances have different semantic similarities to each class, we also extend this relational match to the intra-class level. Our method is simple yet practical, and extensive experiments demonstrate that it adapts well to various architectures, model sizes and training strategies, and can achieve state-of-the-art performance consistently on image classification, object detection, and semantic segmentation tasks. Code is available at: https://github.com/hunto/DIST_KD .