TVT: Training-Free Vision Transformer Search on Tiny Datasets

November 24, 2023 · Declared Dead · 🏛 arXiv.org

⏳ CAUSE OF DEATH: Coming Soon™
Promised but never delivered

"Paper promises code 'coming soon'"

Evidence collected by the PWNC Scanner

Authors Zimian Wei, Hengyue Pan, Lujun Li, Peijie Dong, Zhiliang Tian, Xin Niu, Dongsheng Li arXiv ID 2311.14337 Category cs.CV: Computer Vision Citations 10 Venue arXiv.org Last Checked 1 month ago
Abstract
Training-free Vision Transformer (ViT) architecture search is presented to search for a better ViT with zero-cost proxies. While ViTs achieve significant distillation gains from CNN teacher models on small datasets, the current zero-cost proxies in ViTs do not generalize well to the distillation training paradigm according to our experimental observations. In this paper, for the first time, we investigate how to search in a training-free manner with the help of teacher models and devise an effective Training-free ViT (TVT) search framework. Firstly, we observe that the similarity of attention maps between ViT and ConvNet teachers affects distill accuracy notably. Thus, we present a teacher-aware metric conditioned on the feature attention relations between teacher and student. Additionally, TVT employs the L2-Norm of the student's weights as the student-capability metric to improve ranking consistency. Finally, TVT searches for the best ViT for distilling with ConvNet teachers via our teacher-aware metric and student-capability metric, resulting in impressive gains in efficiency and effectiveness. Extensive experiments on various tiny datasets and search spaces show that our TVT outperforms state-of-the-art training-free search methods. The code will be released.
Community shame:
Not yet rated
Community Contributions

Found the code? Know the venue? Think something is wrong? Let us know!

📜 Similar Papers

In the same crypt — Computer Vision

Died the same way — ⏳ Coming Soon™