Knowledge Distillation via the Target-aware Transformer

TL;DR

This paper introduces a target-aware transformer for knowledge distillation that improves the transfer of semantic information by allowing each teacher pixel to influence all student spatial locations, outperforming previous methods.

Contribution

It proposes a novel one-to-all spatial matching approach using a target-aware transformer for more effective knowledge distillation.

Findings

Surpasses state-of-the-art on ImageNet, Pascal VOC, and COCOStuff10k.

Significantly improves small neural network performance.

Demonstrates the effectiveness of one-to-all spatial matching in knowledge transfer.

Abstract

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.