AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Rethinking Mobile Block for Efficient Attention-based Models

Jiangning Zhang, Xiangtai Li, Jian Li, Liang Liu, Zhucun Xue, Boshen Zhang, Zhengkai Jiang, Tianxin Huang, Yabiao Wang, Chengjie Wang

ICCV 2023 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 360 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 DW-Conv to iRMB (EW-MHSA only vs both) (ablation)EMO-5M · ImageNet-1K valTop-1 accuracy: 76.6 -> 78.4Improved
add EW-MHSA to iRMB (DW-Conv only vs both) (ablation)EMO-5M · ImageNet-1K valTop-1 accuracy: 77.6 -> 78.4Improved
replace Transformer blocks with iRMBDeiT-Tiny · ImageNet-1K valTop-1 accuracy: 72.2 -> 74.3Improved
replace Transformer blocks with iRMBPVT-Tiny · ImageNet-1K valTop-1 accuracy: 75.1 -> 75.4Improved
use one-residual iRMB instead of two-residual MetaFormer with iRMB token mixer (ablation)EMO-5M · ImageNet-1K valTop-1 accuracy: 77.5 -> 78.4Improved
cascade DW-Conv and EW-MHSA instead of parallel (ablation)EMO-5M · ImageNet-1K valTop-1 accuracy: 78.1 -> 78.4Improved
put EW-MHSA before DW-Conv rather than switching order (ablation)EMO-5M · ImageNet-1K valTop-1 accuracy: -0.6 when order switchedImproved
use stronger training recipes (DeiT, EdgeNeXt) instead of simple recipe (ablation)EMO-5M · ImageNet-1K valTop-1 accuracy: 78.4 vs 78.1 / 78.3No clear effect
apply MHSA in all four stages instead of last two (ablation)EMO-5M · ImageNet-1K valTop-1 accuracy / FLOPs: 78.4 -> 78.8, FLOPs 903M -> 992MMixed
use LN in all stages instead of stages 3/4 (ablation)EMO-5M · ImageNet-1K valthroughput: 1,693 -> 952 throughput, +0.2 Top-1Hurt
use EMO block instead of EdgeNeXt block for mobile inferenceEMO-5M · iPhone14 latency (ms)latency: 27.7 -> 6.8Improved
replace operators with group conv, asymmetric conv or performer (ablation)iRMB · ImageNet-1K valTop-1 accuracyNo 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