AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

PVMamba: Parallelizing Vision Mamba via Dynamic State Aggregation

Fei Xie, Zhongdao Wang, Weijia Zhang, Chao Ma

ICCV 2025 · Architectures and efficiency › Attention and transformers · approach: Mamba architectures · 3 citations (OpenAlex)

Read the paper

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
replace SS2D with Sa-DSA operator (ablation)Mamba baseline · ImageNet-1KTop-1 accuracy (%): 82.5 -> 83.0Improved
add hybrid model stage design (ablation)PVMamba tiny · ImageNet-1KTop-1 accuracy (%): 83.0 -> 83.2Improved
add MESA (ablation)PVMamba tiny · ImageNet-1KTop-1 accuracy (%): 83.5 -> 83.9Improved
use Lo-DSA in the first two stages instead of Sa-DSA (ablation)PVMamba lite · ImageNet-1KTop-1 accuracy (%): 82.0 -> 82.2Improved
use Lo-DSA in first two stages instead of Sa-DSA (ablation)PVMamba lite · ImageNet-1KThroughput (img/s): 1128 -> 1423Improved
replace stage 4 DSA with vanilla attention (ablation)PVMamba lite · ImageNet-1KTop-1 accuracy (%): 82.2 -> 82.2No clear effect
replace stage 4 DSA with vanilla attention (ablation)PVMamba lite · ImageNet-1KThroughput (img/s): 1423 -> 1524Improved
increase sampling points Kd from 3 to 5 (ablation)PVMamba lite · ImageNet-1KTop-1 accuracy (%): 82.1 -> 82.2Mixed
increase sampling points Kd from 3 to 9 (ablation)PVMamba lite · COCObox AP: 45.2 -> 45.7Improved
increase sampling points Kd from 5 to 9 (ablation)PVMamba lite · ImageNet-1KMemory cost: 5.8GB -> 6.5GBHurt
add layer multiplexing in interleaved pattern B (ablation)Sa-DSA stack · ImageNet-1Ktraining stabilityImproved
replace sequential SSM with parallel DSA, no acceleration (ablation)SSM · latencyLatency (ms): 0.613 -> 0.702Mixed

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