ViT-AdaLA: Adapting Vision Transformers with Linear Attention

TL;DR

ViT-AdaLA introduces a three-stage framework to adapt and transfer knowledge from vision foundation models to linear attention vision transformers, improving scalability and performance on vision tasks.

Contribution

It proposes a novel three-stage method for effectively adapting and transferring prior knowledge from VFMs to linear attention ViTs, addressing existing limitations.

Findings

Outperforms state-of-the-art linear attention models on classification tasks.

Demonstrates improved segmentation performance with the adapted ViT-AdaLA.

Effective in reducing computational complexity while maintaining accuracy.

Abstract

Vision Transformers (ViTs) based vision foundation models (VFMs) have achieved remarkable performance across diverse vision tasks, but suffer from quadratic complexity that limits scalability to long sequences. Existing linear attention approaches for ViTs are typically trained from scratch, requiring substantial computational resources, while linearization-based methods developed for large language model decoders do not transfer well to ViTs. To address these challenges, we propose ViT-AdaLA, a novel framework for effectively adapting and transferring prior knowledge from VFMs to linear attention ViTs. ViT-AdaLA consists of three stages: attention alignment, feature alignment, and supervised fine-tuning. In the attention alignment stage, we align vanilla linear attention with the original softmax-based attention in each block to approximate the behavior of softmax attention. However, residual approximation errors inevitably accumulate across layers. We mitigate this by fine-tuning the linearized ViT to align its final-layer features with a frozen softmax VFM teacher. Finally, the adapted prior knowledge is transferred to downstream tasks through supervised fine-tuning. Extensive experiments on classification and segmentation tasks demonstrate the effectiveness and generality of ViT-AdaLA over various state-of-the-art linear attention counterpart.