AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

SSM-Aware Token-Efficient VMamba via Adaptive Patch Pruning and Merging for Person Re-Identification

Huiyuan Huang, Sang Min Yoon

CVPR 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Layer-wise pruning of LLMs

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 Delta-guided token pruning to VMamba (ablation)VMamba-Base · Market-1501FLOPs (G) / Rank-1: 15.46 -> 3.83 FLOPs; Rank-1 98.2 -> 96.7Mixed
add token merging only (no pruning) (ablation)VMamba-Base · Market-1501Rank-1 / mAP: 98.2 -> 98.5 Rank-1; 91.6 -> 92.1 mAPImproved
add token merging only (no pruning) (ablation)VMamba-Tiny · Market-1501mAP: 91.6 -> 91.1No clear effect
add merging on top of pruning (ablation)VMamba-Base with pruning · Market-1501Rank-1 / FLOPs: 96.7 -> 96.7 Rank-1; 3.83 -> 3.82 FLOPsNo clear effect
apply pruning on smaller backbone (Tiny) instead of BaseTE-VMamba · Market-1501Rank-1: -4.6 Rank-1 (Tiny) vs -1.5 (Base)Hurt
order pruning before merging instead of merging before pruning (ablation)TE-VMamba-Base · Market-1501FLOPs (G): 7.69 -> 3.82Improved
order pruning before merging instead of applying both simultaneously (ablation)TE-VMamba-Base · Market-1501Rank-1: 95.5 -> 96.7Improved
use dot-product similarity instead of L2 distance for merging (ablation)TE-VMamba-Base · Market-1501Rank-1: 95.5 -> 96.7Improved
use cosine or kNN or RBF similarity instead of dot-product (ablation)TE-VMamba-Base · Market-1501mAP: 87.1 -> 87.0 (cosine)No clear effect
apply token adaptation at inference to query and gallery (ablation)TE-VMamba-Base · CUHK03-NP (Detected)Rank-1 / FLOPs (G): 94.6 -> 95.4 Rank-1; 15.46 -> 3.82 FLOPsImproved
apply token adaptation at inference to query and gallery (ablation)TE-VMamba-Base · Market-1501Rank-1 / throughput: 98.2 -> 96.7 Rank-1; 322.42 -> 590.07 im/sMixed

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