CVMar 18, 2022

Three things everyone should know about Vision Transformers

arXiv:2203.09795v1173 citationsh-index: 60
Originality Incremental advance
AI Analysis

This work addresses efficiency and adaptability challenges in vision transformers for computer vision researchers and practitioners, offering incremental improvements.

The paper tackles improving vision transformers by proposing three simple modifications: parallel processing of residual layers, fine-tuning only attention layers for adaptation, and adding MLP-based patch pre-processing for self-supervised training, achieving efficient performance with reduced compute and memory usage.

After their initial success in natural language processing, transformer architectures have rapidly gained traction in computer vision, providing state-of-the-art results for tasks such as image classification, detection, segmentation, and video analysis. We offer three insights based on simple and easy to implement variants of vision transformers. (1) The residual layers of vision transformers, which are usually processed sequentially, can to some extent be processed efficiently in parallel without noticeably affecting the accuracy. (2) Fine-tuning the weights of the attention layers is sufficient to adapt vision transformers to a higher resolution and to other classification tasks. This saves compute, reduces the peak memory consumption at fine-tuning time, and allows sharing the majority of weights across tasks. (3) Adding MLP-based patch pre-processing layers improves Bert-like self-supervised training based on patch masking. We evaluate the impact of these design choices using the ImageNet-1k dataset, and confirm our findings on the ImageNet-v2 test set. Transfer performance is measured across six smaller datasets.

Code Implementations8 repos

Data from Papers with Code (CC-BY-SA-4.0)

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes