AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

ViTAEv2: Vision Transformer Advanced by Exploring Inductive Bias for Image Recognition and Beyond

Qiming Zhang, Yufei Xu, Jing Zhang, Dacheng Tao

IJCV 2023 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 487 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 PCM to reduction cell (ablation)ViTAE-T · ImageNet-1K (100 ep)top-1 accuracy: 69.8 -> 71.7Improved
add pre fusion and BN in normal cell on top of PCM (ablation)ViTAE-T · ImageNet-1K (100 ep)top-1 accuracy: 71.7 -> 72.6Improved
add BN to pre fusion normal cell (ablation)ViTAE-T · ImageNet-1K (100 ep)top-1 accuracy: 69.1 -> 69.9Improved
raise dilation set to [1,2,3,4,5] (ablation)ViTAE-T · ImageNet-1K (100 ep)top-1 accuracy: 69.9 -> 68.9Hurt
use smaller dilation rates in deeper reduction cells (ablation)ViTAE-T · ImageNet-1K (100 ep)top-1 accuracy: 69.2 -> 69.8Improved
replace Performer attention with window attention in stage 1 (ablation)ViTAEv2 · ImageNet-1Ktop-1 accuracy: 82.2 -> 82.7Improved
use window attention in all four stages instead of first two (ablation)ViTAEv2 · ImageNet-1Ktop-1 accuracy: 82.6 -> 82.2Hurt
add window shifting and relative position encoding (ablation)ViTAEv2-S · ImageNet-1Ktop-1 accuracy: 82.6 -> 82.7No clear effect
add 1x1 convolution branch during pretraining (ablation)ViTAE-L (MAE) · ImageNet-1Ktop-1 accuracy: 84.1 -> 84.6Improved
add 3x3 convolution branch during pretraining (ablation)ViTAE-L (MAE) · ImageNet-1Ktop-1 accuracy: 84.1 -> 84.4Improved
replace Swin-T with ViTAEv2-S backboneSwin-T · COCO Mask R-CNN 1xbox AP: 43.7 -> 46.3Improved
replace Swin-T with ViTAEv2-S backboneSwin-T · ImageNet-1Ktop-1 accuracy: 81.3 -> 82.6Improved

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