AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Going Deeper With Image Transformers

Hugo Touvron, Matthieu Cord, Alexandre Sablayrolles, Gabriel Synnaeve, Hervé Jégou

ICCV 2021 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 1,425 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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
add LayerScale (per-channel residual scaling) to a 36-block model (ablation)DeiT-S depth 36 · ImageNet-1k valtop-1 accuracy: 81.9 -> 82.9Improved
use per-channel LayerScale instead of ReZero scalar weighting (ablation)DeiT-S depth 36 · ImageNet-1k valtop-1 accuracy: 81.6 -> 82.9Improved
initialize LayerScale at small eps instead of 0 (ablation)DeiT-S depth 36 · ImageNet-1k valtop-1 accuracy: 82.7 -> 82.9Improved
initialize LayerScale at constant eps instead of uniform random U[0,2eps] (ablation)DeiT-S depth 36 · ImageNet-1k valtop-1 accuracy: 82.6 -> 82.9Improved
keep scale factors learnable instead of fixed to values from a previous LayerScale run (ablation)DeiT-S depth 36 · ImageNet-1k valtop-1 accuracy: 81.6 -> 82.9Improved
adapt stochastic depth drop rate when adding LayerScale (ablation)DeiT-S depth 36 · ImageNet-1k valtop-1 accuracy: 80.5 -> 83.0Improved
add class-attention layers (12 SA + 2 CA) to DeiT-S (ablation)DeiT-S · ImageNet-1k valtop-1 accuracy: 79.9 -> 80.8Improved
insert the class token late, at layer 10 instead of layer 0 (ablation)DeiT-S · ImageNet-1k valtop-1 accuracy: 79.9 -> 80.5Improved
add class-attention architecture on top of LayerScale (ablation)CaiT-S36 · ImageNet-1k valtop-1 accuracy: 83.0 -> 83.2Improved
exclude the class embedding from keys of class-attention (ablation)CaiT-S36 · ImageNet-1k valtop-1 accuracy: 83.44 -> 83.31No clear effect
remove LayerScale in the class-attention blocks (ablation)CaiT-S36 · ImageNet-1k valtop-1 accuracy: 83.36 -> 83.44No clear effect
add a distillation token to hard distillation (ablation)CaiT · ImageNet-1k valtop-1 accuracy: 78.4 -> 78.5 (XXS-24); 84.8 -> 84.7 (M-24)No clear effect

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