Awesome 3D Token Compression [](https://awesome.re)
August 19, 2026 · View on GitHub
Awesome 3D Token Compression 
A curated list of token compression methods — pruning · merging · reduction — for 3D multimodal LLMs. ⚡🧊
Keywords: 3D token compression · 3D token pruning · 3D token reduction · 3D token merging · efficient 3D LLM · 3D VLM · 3D scene understanding
3D multimodal LLMs lift multi-view features into world space, producing thousands of tokens per scene — and the LLM backbone spends >90% of its compute on them. This list tracks methods that prune, merge, or compress those 3D visual tokens to make inference efficient without sacrificing spatial understanding.
Important
⭐ Found this useful? Give it a star — it helps others discover the list and keeps it growing.
📝 Add your paper: Open an issue and we'll add it.
Legend · Code: ✅ available · 🔒 not released
📰 News
- 2026-08-19 — Added CoverPrune (ECCV'26 Spotlight), Geo3DPruner, Efficient3D, HCC-3D, and DTC.
- 2026-08-19 — 🎉 Awesome 3D Token Compression is live!
Papers
2026
| Method | Title & Authors | Backbone | Code |
|---|---|---|---|
| 3DZip |   Spatial-Aware Feature Diversity-Guided Token Compression for 3D QA Changwoo Baek et al. | LLaVA-3D · Video-3D LLM · SR-3D | ✅ |
| CoverPrune |  Coverage-Driven Token Pruning for 3D VLMs via Optimal Transport Peng Ling et al. | GS-Reasoner · VLM-3R | 🔒 |
| Merge3D |  Efficient 3D Multimodal LLMs via Joint 2D-3D Token Merging Tianbo Pan et al. | Multi-view 3D MLLM | 🔒 |
| Geo3DPruner |  Geometry-Guided 3D Visual Token Pruning for Video-Language Models Han Li et al. | Video-3D LLM (LLaVA-Video) | 🔒 |
| Efficient3D |   A Unified Framework for Adaptive and Debiased Token Reduction in 3D MLLMs Yuhui Lin et al. | Chat-Scene | ✅ |
| HCC-3D |   Hierarchical Compensatory Compression for 98% 3D Token Reduction in VLMs Liheng Zhang et al. | GreenPLM (Phi-2/3) | ✅ |
2025
| Method | Title & Authors | Backbone | Code |
|---|---|---|---|
| Fast3D |   Accelerating 3D Multi-modal LLMs for Efficient 3D Scene Understanding Wencan Huang et al. | Chat-Scene | ✅ |
| AdaToken-3D |  Dynamic Spatial Gating for Efficient 3D Large Multimodal-Models Reasoning Kai Zhang et al. | LLaVA-3D | 🔒 |
| DTC |  Zero-shot 3D Question Answering via Voxel-based Dynamic Token Compression Hsiang-Wei Huang et al. | Multi-frame VLM | 🔒 |
🤝 Contributing
Contributions are very welcome! To add a paper, open a PR that:
- Places the entry in the correct year section, sorted by venue date (newest first).
- Follows the row format:
**Method** | venue badge + [Title](link) + authors | Backbone | Code. - Adds a venue badge via shields.io — venue name colored, year grey (e.g.
); useif the paper has no venue yet. - Links code with ✅ (or marks 🔒 if unreleased).
📄 License
To the extent possible under law, the contributors have waived all copyright and related rights to this work.