Learning Efficient Vision Transformers via Fine-Grained Manifold Distillation

TL;DR

This paper introduces a fine-grained manifold distillation method for vision transformers that effectively compresses models, leading to high accuracy on ImageNet and improved transfer learning performance, suitable for edge devices.

Contribution

It proposes a novel patch-level manifold distillation technique specifically designed for vision transformers, reducing computational costs while maintaining high accuracy.

Findings

DeiT-Tiny model achieves 76.5% top-1 accuracy on ImageNet-1k.

The method outperforms previous distillation approaches by 2.0%.

Demonstrates superior transfer learning results on various benchmarks.

Abstract

In the past few years, transformers have achieved promising performances on various computer vision tasks. Unfortunately, the immense inference overhead of most existing vision transformers withholds their from being deployed on edge devices such as cell phones and smart watches. Knowledge distillation is a widely used paradigm for compressing cumbersome architectures via transferring information to a compact student. However, most of them are designed for convolutional neural networks (CNNs), which do not fully investigate the character of vision transformer (ViT). In this paper, we utilize the patch-level information and propose a fine-grained manifold distillation method. Specifically, we train a tiny student model to match a pre-trained teacher model in the patch-level manifold space. Then, we decouple the manifold matching loss into three terms with careful design to further reduce the computational costs for the patch relationship. Equipped with the proposed method, a DeiT-Tiny model containing 5M parameters achieves 76.5% top-1 accuracy on ImageNet-1k, which is +2.0% higher than previous distillation approaches. Transfer learning results on other classification benchmarks and downstream vision tasks also demonstrate the superiority of our method over the state-of-the-art algorithms.