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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| apply VLZip unified compression to Qwen2.5-VL-3B | Qwen2.5-VL-3B · LongVLBench (avg) | average score: 41.4 -> 61.9 | Improved |
| apply VLZip at ultra-long input | Qwen2.5-VL-3B · LongVLBench (>128k) | score: 0.16 -> 7.60 | Improved |
| apply VLZip at 128k context | Qwen2.5-VL-3B · MMLongBench ICL 128k | score: 7.5 -> 58.8 | Improved |
| apply VLZip at short contexts | Qwen2.5-VL-3B · MMLongBench VRAG 8k | score: 43.4 -> 25.2 | Hurt |
| apply VLZip visual compression on short-context VQA | Qwen2.5-VL-3B · Short-context VQA (GQA, MMBench, OKVQA) | accuracy: GQA 60.0 -> 53.8; OKVQA 42.5 -> 51.1 | Mixed |
| 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 128k | score: 4.3 -> 58.8 | Improved |
| add Stage 4 long-context consolidation (ablation) | VLZip · MMLongBench ICL 128k | score: 46.8 -> 58.8 | Improved |
| add Stage 1 and 2 compressor pretraining (ablation) | VLZip · LongVLBench (avg) | average score: 58.6 -> 61.9 | Improved |
| increase tokens per chunk from 4 to 8 (ablation) | VLZip · MMLongBench ICL 128k | score: 58.8 -> 20.5 | Hurt |
| decrease tokens per chunk from 4 to 2 (ablation) | VLZip · MMLongBench ICL 128k | score: 58.8 -> 39.3 | Hurt |
| reduce chunk size from 100 to 25 (ablation) | VLZip · VRAG/NIAH avg vs ICL-128k | score: VRAG avg 26.0 -> 30.7; ICL-128k 58.8 -> 12.8 | Mixed |
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.
- VisionZip — VisionZip: Longer is Better but Not Necessary in Vision Language Models (CVPR 2025) · better on 1 of 1 shared result
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
- Hybrid Token Compression for Vision-Language Models (CVPR 2026 · similarity 0.91)
- Efficient Large Multi-modal Models via Visual Context Compression (NeurIPS 2024 · similarity 0.90)
- LLMC+: Benchmarking Vision-Language Model Compression with a plug-and-play Toolkit (AAAI 2026 · similarity 0.90)
- DUET-VLM: Dual stage Unified Efficient Token reduction for VLM Training and Inference (CVPR 2026 · similarity 0.89)
- ApET: Approximation-Error Guided Token Compression for Efficient VLMs (CVPR 2026 · similarity 0.89)
- Accelerating Multimodal Large Language Models via Dynamic Visual-Token Exit and the Empirical Findings (NeurIPS 2025 · similarity 0.89)
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.
