Knowledge Distillation via the Target-aware Transformer
This work improves knowledge distillation for computer vision tasks, offering a novel method to enhance small network performance, though it is incremental as it builds on existing distillation techniques.
The paper tackles the problem of knowledge distillation for small neural networks by addressing the mismatch in semantic information between teacher and student features due to architectural differences, proposing a one-to-all spatial matching approach using a target-aware transformer that achieves state-of-the-art performance on benchmarks like ImageNet, Pascal VOC, and COCOStuff10k.
Knowledge distillation becomes a de facto standard to improve the performance of small neural networks. Most of the previous works propose to regress the representational features from the teacher to the student in a one-to-one spatial matching fashion. However, people tend to overlook the fact that, due to the architecture differences, the semantic information on the same spatial location usually vary. This greatly undermines the underlying assumption of the one-to-one distillation approach. To this end, we propose a novel one-to-all spatial matching knowledge distillation approach. Specifically, we allow each pixel of the teacher feature to be distilled to all spatial locations of the student features given its similarity, which is generated from a target-aware transformer. Our approach surpasses the state-of-the-art methods by a significant margin on various computer vision benchmarks, such as ImageNet, Pascal VOC and COCOStuff10k. Code is available at https://github.com/sihaoevery/TaT.