AI Paper Atlas

Research map / Video / Video-language understanding

Perceptual Projection Pruning: Diversity-Aware Video Token Pruning for Multimodal Large Language Models

Zhuangqiu Huang, Minxin Lai, Shuo Liu, Yu Zhang, Jiaqi Wang

ECCV 2026 · Video › Video-language understanding · approach: Video LLMs with fine-grained understanding · 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
cluster on 3D saliency instead of high-dim raw features (ablation)P3 (adaptive) · 4 video benchmarks avg, 25% retentionaverage delta: +0.10 for raw featuresNo clear effect
add dynamic saliency (SM) (ablation)P3 · 4 video benchmarks avg, 25% retentionaverage delta: -1.40 when removedImproved
add regional saliency (SR) (ablation)P3 · 4 video benchmarks avg, 25% retentionaverage delta: -0.75 when removedImproved
add focal saliency (SF) (ablation)P3 · 4 video benchmarks avg, 25% retentionaverage delta: -0.43 when removedImproved
use K-Means instead of DPC-KNN (ablation)P3 · 4 video benchmarks avg, 25% retentionaverage: 62.1 -> 64.3Improved
uniform temporal step 4 instead of step 2 (ablation)P3 · 4 video benchmarks avg, 25% retentionaverage: 64.2 -> 65.2Improved
adaptive instead of uniform partitioning on multi-turn (ablation)P3 · MT-Video-Benchaccuracy: 62.5 -> 63.0Improved
vary cluster count K from 5 to 80 (ablation)P3 · 4 video benchmarksaccuracyNo clear effect
query-agnostic P3 vs query-aware CDPruner at 25% retentionP3 · Qwen3-VL-8B 4 video benchmarksavg retained (%): 98.1 -> 97.3Hurt
P3 on query-dependent temporal localizationP3 · Charades-STAscore: 47.9 (CDPruner) -> 44.8Hurt
P3 on spatial reasoningP3 · VSI-Benchaccuracy: 55.4 (vanilla) -> 59.5Improved
P3 for 1024-frame inputsP3 · 1024-frame, 25% retentiontotal latency (s): 19.57s prefill -> 2.22s totalImproved

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