AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

ToDRE: Effective Visual Token Pruning via Token Diversity and Task Relevance

Duo Li, Zuhao Yang, Xiaoqin Zhang, Ling Shao, Shijian Lu

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Attention-based visual token pruning

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
add relevance-driven Stage 2 on top of diversity Stage 1 (ablation)ToDRE LLaVA-NeXT-7B · 4-benchmark avg, 10% retentionaverage relative performance (%): 95.8 -> 96.0Improved
add relevance-driven Stage 2 on top of diversity Stage 1 (ablation)ToDRE LLaVA-NeXT-7B · 4-benchmark avg, 25% retentionaverage relative performance (%): 98.8 -> 98.9Improved
add relevance-driven Stage 2 on top of diversity Stage 1 (ablation)ToDRE LLaVA-NeXT-7B · 4-benchmark total inference time, 10% retentiontotal time (min:sec): 31:18 -> 29:43Improved
apply Stage 2 alone to the unpruned model (ablation)LLaVA-NeXT-7B · 4-benchmark avgtotal time (min:sec): 77:04 -> 70:15Improved
select pivot by [CLS] attention rather than random (ablation)ToDRE Stage 1 · 4-benchmark avg, 10% retentionaverage relative performance (%): 95.8 -> 96.0Improved
select pivot by [CLS] attention rather than center token (ablation)ToDRE Stage 1 · 4-benchmark avg, 10% retentionaverage relative performance (%): 95.6 -> 96.0Improved
prune at layer 7L/8 instead of 5L/8 (ablation)ToDRE Stage 2 · 4-benchmark avgaverage relative performance (%): 96.2 -> 100.0Improved
prune at layer 7L/8 instead of 6L/8 (ablation)ToDRE Stage 2 · 4-benchmark avgaverage relative performance (%): 98.3 -> 100.0Improved
prune at layer 7L/8 instead of L/2 (ablation)ToDRE Stage 2 · 4-benchmark avgaverage relative performance (%): 87.9 -> 100.0Improved
set threshold tau=0.10 instead of 0.03 (ablation)ToDRE Stage 2 · 4-benchmark avgtotal time (min:sec): 79:22 -> 72:35Improved
vary threshold tau from 0.05 to 0.15 (ablation)ToDRE Stage 2 · 4-benchmark avgaverage relative performance (%): 100.0 / 100.1 / 100.0No clear effect
scale ToDRE to LLaVA-NeXT-13B vs DivPruneToDRE · 8 image benchmarks avg, 10% retentionaverage relative performance (%): 92.5 -> 93.6Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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