AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Partial Ring Scan: Revisiting Scan Order in Vision State Space Models

Yi-Kuan Hsieh, Kuan-Chuan Peng, Xin Li, Ming-Ching Chang, Yu-Chee Tseng, Jun Wei Hsieh

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: State space and recurrent sequence models · 0 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 partial channel filtering (PCF) to PRIS-Mamba (ablation)PRIS-Mamba · ImageNet-1K valTop-1 accuracy (%): 84.1 -> 84.5Improved
add PCF to PRIS-Mamba (ablation)PRIS-Mamba · ImageNet-1K valthroughput (img/s): 2177 -> 2854Improved
plug PCF into VMamba (ablation)VMamba · ImageNet-1K valTop-1 accuracy (%): 82.6 -> 82.9Improved
plug PCF into other Vision-Mamba backbones (Vim, QuadMamba, Adventurer, SparX, GroupMamba) (ablation)Vision-Mamba backbones · ImageNet-1K valTop-1 accuracy (%): +0.2 to +0.4 ppImproved
replace hard PCF with soft channel attention SE (ablation)PRIS-Mamba w/o PCF · ImageNet-1K valTop-1 accuracy (%): 84.1 -> 84.2; throughput 2177 -> 2089No clear effect
add CBAM soft channel attention instead of PCF (ablation)PRIS-Mamba w/o PCF · ImageNet-1K valTop-1 accuracy (%): 84.1 -> 84.3; params 27M -> 30MNo clear effect
use ring scan instead of fixed-path scansPRIS-Mamba · ImageNet-1K val, 60 deg rotationTop-1 accuracy (%): PRIS-Mamba 84.5 -> 84.4 vs VMamba 82.6 -> 80.6Improved
replace ring scan with single or multi-path fixed scans on the same backbone (ablation)PRIS-Mamba-S · ImageNet-1K valTop-1 accuracy (%): 84.9 vs 82.1-84.1 (best four-path S21 84.1)Hurt
add more scan paths (1 to 2 to 4) to fixed-path scan (ablation)fixed-path Vision Mamba · ImageNet-1K rotation 30/60 degTop-1 drop under rotation: losses of ~1.2-1.9 pp at 30/60 deg for all S1-S21No clear effect
vary ring width 1, 2, 4 (ablation)PRIS-Mamba-T · ImageNet-1K valTop-1 accuracy (%): 84.3 / 84.5 / 84.3No clear effect
use ring scan under random-mask occlusionPRIS-Mamba 30M · ImageNet-1K val, mask 16x16Top-1 drop: -0.4 vs VMamba -0.9Improved
use ring scan under patch shufflingPRIS-Mamba 30M · ImageNet-1K val, cut 16 patchesTop-1 drop: -0.7 vs VMamba -1.3Improved

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