AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

UniFormer: Unifying Convolution and Self-Attention for Visual Recognition

Kunchang Li, Yali Wang, Junhao Zhang, Peng Gao, Guanglu Song, Yu Liu, Hongsheng Li, Yu Qiao

TPAMI 2023 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 693 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 Token Labeling (ablation)UniFormer-S · ImageNet-1K valtop-1 accuracy: 82.9 -> 83.4Improved
add Token Labeling (ablation)UniFormer-B · ImageNet-1K valtop-1 accuracy: 83.9 -> 85.1Improved
fine-tune at 384x384 resolutionUniFormer-S with Token Labeling · ImageNet-1K valtop-1 accuracy: 83.4 -> 84.6Improved
use overlapped patch embedding and more blocks (ablation)UniFormer-S · ImageNet-1K valtop-1 accuracy: 82.9 -> 83.4Improved
use global MHRA instead of window-based hybrid in Stage3 (ablation)UniFormer-S · ADE20K Semantic FPNmIoU: 46.2 -> 46.6 mIoU but 199 -> 247 GFLOPsImproved
use global MHRA instead of window-based hybrid in Stage3 (ablation)UniFormer-B · ADE20K Semantic FPNmIoU: 47.7 -> 48.0Improved
use global MHRA instead of window-based hybrid in Stage3 (ablation)UniFormer-S · ADE20K UperNetmIoU: 47.0 -> 47.6Improved
pretrain on Kinetics-600 instead of Kinetics-400UniFormer-S 16 frames · Something-Something V1top-1 accuracy: 53.8 -> 54.4Improved
pretrain on Kinetics-600 instead of Kinetics-400UniFormer-B 16 frames · Something-Something V1top-1 accuracy: 59.1 -> 58.8No clear effect
increase from 16 to 32 framesUniFormer-B · Kinetics-400top-1 accuracy: 82.0 -> 82.9Improved
increase from 1 to 3 spatial cropsUniFormer-B 32 frames · Kinetics-400top-1 accuracy: 82.9 -> 83.0No 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