Timezone: »

Augmented Shortcuts for Vision Transformers
Yehui Tang · Kai Han · Chang Xu · An Xiao · Yiping Deng · Chao Xu · Yunhe Wang

Tue Dec 07 04:30 PM -- 06:00 PM (PST) @ None #None

Transformer models have achieved great progress on computer vision tasks recently. The rapid development of vision transformers is mainly contributed by their high representation ability for extracting informative features from input images. However, the mainstream transformer models are designed with deep architectures, and the feature diversity will be continuously reduced as the depth increases, \ie, feature collapse. In this paper, we theoretically analyze the feature collapse phenomenon and study the relationship between shortcuts and feature diversity in these transformer models. Then, we present an augmented shortcut scheme, which inserts additional paths with learnable parameters in parallel on the original shortcuts. To save the computational costs, we further explore an efficient approach that uses the block-circulant projection to implement augmented shortcuts. Extensive experiments conducted on benchmark datasets demonstrate the effectiveness of the proposed method, which brings about 1% accuracy increase of the state-of-the-art visual transformers without obviously increasing their parameters and FLOPs.

Author Information

Yehui Tang (Peking University)
Kai Han (Huawei Noah's Ark Lab)
Chang Xu (The University of Sydney)
An Xiao (Tsinghua University, Tsinghua University)
Yiping Deng (Tsinghua University, Tsinghua University)
Chao Xu (Peking University)
Yunhe Wang (Huawei Noah's Ark Lab)

More from the Same Authors