AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

LipsFormer: Introducing Lipschitz Continuity to Vision Transformers

Xianbiao Qi, Jianan Wang, Yihao Chen, Yukai Shi, Lei Zhang

ICLR 2023 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 4 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Replace the CSwin-T design with LipsFormer components and train without warmupCSwin-T · ImageNet-1KTop-1 accuracy: 82.7 -> 83.5Improved
Replace Swin-T with LipsFormer-Swin-TSwin-T · ImageNet-1Kaccuracy: 81.3 -> 82.7Improved
Add 5 epochs of learning-rate warmup (ablation)LipsFormer-CSwin-T · ImageNet-1K, 100 epochsTop-1 accuracy: 81.6 -> 81.2Hurt
Add CenterNorm instead of using no normalization (ablation)LipsFormer-CSwin-T · ImageNet-1K, 100 epochsTop-1 accuracy: 80.3 -> 81.6Improved
Replace CenterNorm with LayerNorm (ablation)LipsFormer-CSwin-T · ImageNet-1K, 100 epochsconvergenceHurt
Use spectral initialization instead of truncated-normal initialization (ablation)LipsFormer-CSwin-T · ImageNet-1K, 100 epochsTop-1 accuracy: 81.3 -> 81.6Improved
Replace scaled cosine similarity attention with dot-product attention (ablation)LipsFormer-CSwin-T · ImageNet-1K, 100 epochsconvergenceHurt
Use scaled cosine similarity attention instead of L2 distance attention (ablation)LipsFormer-CSwin-T · ImageNet-1K, 100 epochsTop-1 accuracy: 81.3 -> 81.6Improved
Increase residual weight alpha from 0.1 to 0.2 (ablation)LipsFormer-CSwin-T · ImageNet-1K, 100 epochsTop-1 accuracy: 81.4 -> 81.6Improved
Increase residual weight alpha from 0.2 to 0.4 (ablation)LipsFormer-CSwin-T · ImageNet-1K, 100 epochsconvergenceHurt
Add convolution blocks, from none to one depth-wise convolution (ablation)LipsFormer-CSwin-T · ImageNet-1K, 100 epochsTop-1 accuracy: 80.7 -> 81.2Improved
Add a point-wise convolution after one depth-wise convolution (ablation)LipsFormer-CSwin-T · ImageNet-1K, 100 epochsTop-1 accuracy: 81.2 -> 81.6Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy