# CoSAG: Compact Semantic Anchor Gaussians via Training-Free Rate-Distortion Coding

arXiv: https://arxiv.org/abs/2607.10237 · HTML: https://mlsanity.com/paper/2607.10237/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 21 of its 25 references that are in the map.
- First posted on arXiv 11 Jul 2026.

## Builds on (most cited first, 21 of 21)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [3D Gaussian Splatting for Real-Time Radiance Field Rendering](https://mlsanity.com/paper/2308.04079/) (2023)
- [Scalability in Perception for Autonomous Driving: Waymo Open Dataset](https://mlsanity.com/paper/1912.04838/) (2019)
- [The Replica Dataset: A Digital Replica of Indoor Spaces](https://mlsanity.com/paper/1906.05797/) (2019)
- [Scaffold-GS: Structured 3D Gaussians for View-Adaptive Rendering](https://mlsanity.com/paper/2312.00109/) (2023)
- [OpenScene: 3D Scene Understanding with Open Vocabularies](https://mlsanity.com/paper/2211.15654/) (2022)
- [Gaussian Grouping: Segment and Edit Anything in 3D Scenes](https://mlsanity.com/paper/2312.00732/) (2023)
- [ConceptFusion: Open-set Multimodal 3D Mapping](https://mlsanity.com/paper/2302.07241/) (2023)
- [HAC: Hash-grid Assisted Context for 3D Gaussian Splatting Compression](https://mlsanity.com/paper/2403.14530/) (2024)
- [Language Embedded 3D Gaussians for Open-Vocabulary Scene Understanding](https://mlsanity.com/paper/2311.18482/) (2023)
- [OpenGaussian: Towards Point-Level 3D Gaussian-based Open Vocabulary Understanding](https://mlsanity.com/paper/2406.02058/) (2024)
- [Weakly Supervised 3D Open-vocabulary Segmentation](https://mlsanity.com/paper/2305.14093/) (2023)
- [ContextGS: Compact 3D Gaussian Splatting with Anchor Level Context Model](https://mlsanity.com/paper/2405.20721/) (2024)
- [Dr. Splat: Directly Referring 3D Gaussian Splatting via Direct Language Embedding Registration](https://mlsanity.com/paper/2502.16652/) (2025)
- [LangSplatV2: High-dimensional 3D Language Gaussian Splatting with 450+ FPS](https://mlsanity.com/paper/2507.07136/) (2025)
- [LUDVIG: Learning-Free Uplifting of 2D Visual Features to Gaussian Splatting Scenes](https://mlsanity.com/paper/2410.14462/) (2024)
- [InstanceGaussian: Appearance-Semantic Joint Gaussian Representation for 3D Instance-Level Perception](https://mlsanity.com/paper/2411.19235/) (2024)
- [GaussianDWM: 3D Gaussian Driving World Model for Unified Scene Understanding and Multi-Modal Generation](https://mlsanity.com/paper/2512.23180/) (2025)
- [GAGS: Granularity-Aware Feature Distillation for Language Gaussian Splatting](https://mlsanity.com/paper/2412.13654/) (2024)
- [CF3: Compact and Fast 3D Feature Fields](https://mlsanity.com/paper/2508.05254/) (2025)
- [LightSplat: Fast and Memory-Efficient Open-Vocabulary 3D Scene Understanding in Five Seconds](https://mlsanity.com/paper/2603.24146/) (2026)

## Built on by (most cited first, 1 of 1)

- [LangStreet: Persistent Language Fields for Anchor-Decoded Street Gaussians](https://mlsanity.com/paper/2609.11616/) (2026)

## Abstract (arXiv)

Open-vocabulary 3D scene understanding is commonly achieved by embedding 2D vision-language features such as CLIP into a 3D Gaussian Splatting scene, turning it into a text-queryable semantic field. However, attaching a high-dimensional feature to each of millions of Gaussians inflates a single scene to gigabytes, which makes storage and deployment the real bottleneck of these fields. Existing compact methods each learn and ship a per-scene codec, an autoencoder, a quantized codebook, or a distilled feature field, entangling field construction with field storage and never compressing the per-Gaussian assignment that holds the bulk of the cost. We argue that construction and storage should be decoupled, and that storage is a rate-distortion problem over the per-Gaussian binding to a small anchor table, a structure no prior open-vocabulary method compresses. We present CoSAG, which constructs the field without any per-scene training through a closed-form transmittance-weighted lift, spatially grounded semantic anchors, and multi-view denoising, and stores it with a spatially predictive entropy coder that ships no decoder. Because the anchors are spatially grounded, the binding is predictable and therefore highly compressible. The transmittance-weighted lift and multi-view denoising yield a clean, view-consistent assignment, so the entropy coder spends almost no rate on correcting noise and instead codes only the residual against its spatial prediction. CoSAG reaches sub-megabyte storage while matching or exceeding the state of the art across the 2D-rendered, 3D-selection, and dense-LSeg protocols, reducing field size by 37 to 76x relative to LangSplatV2 at higher accuracy.