Research map / Vision-language and multimodal / Multimodal LLMs
TokenPacker: Efficient Visual Projector for Multimodal LLM
Wentong Li, Yuqian Yuan, Jian Liu, Dongqi Tang, Song Wang, Jie Qin, Jianke Zhu, Lei Zhang
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 |
|---|---|---|---|
| add region-to-point injection module (ablation) | 2x downsampled baseline projector · VQAv2 | accuracy: 76.4 -> 77.5 | Improved |
| add region-to-point injection module (ablation) | 2x downsampled baseline projector · TextVQA | accuracy: 55.3 -> 56.5 | Improved |
| use learnable query instead of downsampled feature query (ablation) | TokenPacker injection · GQA | accuracy: 61.6 -> 59.8 | Hurt |
| add multi-level features as keys and values (ablation) | TokenPacker injection · TextVQA | accuracy: 56.5 -> 57.2 | Improved |
| add dynamic image slicing (ablation) | TokenPacker · TextVQA | accuracy: 57.2 -> 64.0 | Improved |
| add separator tokens (comma and newline) between grids (ablation) | TokenPacker with image slicing · TextVQA | accuracy: 58.3 -> 64.0 | Improved |
| use dynamic slicing instead of FixedSplit (ablation) | TokenPacker-HD · OCRBench | score: 327 -> 336 | Improved |
| use dynamic slicing instead of AdaptiveSplit (ablation) | TokenPacker-HD · TextVQA | accuracy: 63.4 -> 64.0 | Improved |
| use multi-level layers 12-16-22-23 instead of single layer 23 (ablation) | TokenPacker injection · TextVQA | accuracy: 56.5 -> 57.2 | Improved |
| use shallower multi-level layers 8-12-22-23 (ablation) | TokenPacker injection · VQAv2 | accuracy: 77.9 -> 76.8 | Hurt |
| replace MLP with TokenPacker at 144 tokens | LLaVA-1.5 Vicuna-7B · Avg of six benchmarks | average score: 62.0 -> 62.8 | Improved |
| replace LDP-v2 with TokenPacker at 144 tokens (ablation) | LLaVA-1.5 Vicuna-7B · Avg of six benchmarks | average score: 61.2 -> 62.8 | Improved |
Closest papers by meaning
- A More Word-like Image Tokenization for MLLMs (CVPR 2026 · similarity 0.91)
- Accelerating Multimodal Large Language Models via Dynamic Visual-Token Exit and the Empirical Findings (NeurIPS 2025 · similarity 0.90)
- What Do Visual Tokens Really Encode? Uncovering Sparsity and Redundancy in Multimodal Large Language Models (CVPR 2026 · similarity 0.89)
- Honeybee: Locality-enhanced Projector for Multimodal LLM (CVPR 2024 · similarity 0.89)
- Rethinking Visual Token Reduction in LVLMs Under Cross-Modal Misalignment (AAAI 2026 · similarity 0.89)
- Improving Visual Token Reduction via Rectifying Distortions for Efficient Multimodal LLM Inference (ICML 2026 · 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.
