AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

VLZip: Unified Visual and Textual Compression for Interleaved Long-Context Modeling

Yuqi Zhang, Cheng Chen, Yuyu Guo, Wenjie Yang, Lingchen Meng, Peng Di, Hang Yu, Zuxuan Wu, Yu-Gang Jiang

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Attention-based visual token pruning · 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
apply VLZip unified compression to Qwen2.5-VL-3BQwen2.5-VL-3B · LongVLBench (avg)average score: 41.4 -> 61.9Improved
apply VLZip at ultra-long inputQwen2.5-VL-3B · LongVLBench (>128k)score: 0.16 -> 7.60Improved
apply VLZip at 128k contextQwen2.5-VL-3B · MMLongBench ICL 128kscore: 7.5 -> 58.8Improved
apply VLZip at short contextsQwen2.5-VL-3B · MMLongBench VRAG 8kscore: 43.4 -> 25.2Hurt
apply VLZip visual compression on short-context VQAQwen2.5-VL-3B · Short-context VQA (GQA, MMBench, OKVQA)accuracy: GQA 60.0 -> 53.8; OKVQA 42.5 -> 51.1Mixed
compress both modalities instead of one (ablation)VLZip · Max trainable length (8xA100)K tokens: 20 -> 120 (visual only 50, text only 40)Improved
add Stage 3 joint interleaved fine-tuning (ablation)VLZip · MMLongBench ICL 128kscore: 4.3 -> 58.8Improved
add Stage 4 long-context consolidation (ablation)VLZip · MMLongBench ICL 128kscore: 46.8 -> 58.8Improved
add Stage 1 and 2 compressor pretraining (ablation)VLZip · LongVLBench (avg)average score: 58.6 -> 61.9Improved
increase tokens per chunk from 4 to 8 (ablation)VLZip · MMLongBench ICL 128kscore: 58.8 -> 20.5Hurt
decrease tokens per chunk from 4 to 2 (ablation)VLZip · MMLongBench ICL 128kscore: 58.8 -> 39.3Hurt
reduce chunk size from 100 to 25 (ablation)VLZip · VRAG/NIAH avg vs ICL-128kscore: VRAG avg 26.0 -> 30.7; ICL-128k 58.8 -> 12.8Mixed

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