Is Attentional Channel Processing Design Required? Comprehensive Analysis Of Robustness Between Vision Transformers And Fully Attentional Networks
This work addresses the need for robustness analysis in vision models for researchers and practitioners, though it is incremental as it builds on existing testing frameworks.
The study compared the robustness of traditional Vision Transformers and fully attentional networks (FAN) on ImageNet, finding that FAN models showed improved resistance to white-box attacks but similar transferability in black-box attacks.
The robustness testing has been performed for standard CNN models and Vision Transformers, however there is a lack of comprehensive study between the robustness of traditional Vision Transformers without an extra attentional channel design and the latest fully attentional network(FAN) models. So in this paper, we use the ImageNet dataset to compare the robustness of fully attentional network(FAN) models with traditional Vision Transformers to understand the role of an attentional channel processing design using white box attacks and also study the transferability between the same using black box attacks.