IEEE researchers unveil GenSplatCodec to cut 3D data storage by 17x
Researchers have proposed GenSplatCodec, a unified dual-stream codec that reformulates 3D Gaussian Splatting compression as geometry-guided generative decoding. The system demonstrates a 17x reduction in storage costs compared to existing feed-forward methods while maintaining rendering fidelity for immersive media applications.
Key Takeaways
- GenSplatCodec achieves a 17x reduction in storage costs compared to the state-of-the-art YoNoSplat method.
- Dual-stream architecture splits data into a compact 3D structural stream and a lightweight reference appearance stream.
- The system uses geometry-guided one-step generative decoding to reconstruct high-frequency textures removed during compression.
- Standardized testing on DL3DV and RealEstate10K datasets confirms superior rate-distortion performance over cascaded compression baselines.
Why It Matters
Immersive media applications like AR and virtual twins face a bottleneck in the massive data footprints required for photorealistic 3D scene reconstruction. GenSplatCodec solves this by acknowledging that low-bitrate streams cannot explicitly carry every visual detail; instead, it uses generative AI to 'fill in the blanks' based on a sparse geometric foundation. This shift from pure data recovery to guided synthesis allows high-quality 3D video to be transmitted over consumer-grade bandwidth, moving volumetric content closer to mainstream streaming viability. Watch for whether this generative approach is adopted by standard-setting bodies like JPEG or MPEG for next-tier spatial media codecs.
Additional Context
The release of GenSplatCodec arrives as 3D Gaussian Splatting (3DGS) transitions from a research novelty to a standardized industry tool. In early 2026, the Khronos Group announced the KHR_gaussian_splatting extension for glTF 2.0, a major milestone supported by Google, NVIDIA, and Apple to enable cross-platform 3DGS support. Per theFuture3D (April 2026), this standardization eliminates the fragmentation caused by proprietary file formats, allowing splats to be treated as first-class citizens alongside traditional meshes in digital twin and film production pipelines. Simultaneously, the Alliance for OpenUSD (AOUSD) has integrated 3DGS under its Particle Fields schema, ensuring interoperability within the NVIDIA Omniverse and Adobe ecosystems.
Industrial adoption is also accelerating in enterprise sectors. In May 2026, Esri integrated cloud-based GS generation into ArcGIS Reality, while software providers like PIX4D and Veesus added native 3DGS support for CAD and geospatial workflows. According to IEEE 2026 technology predictions, the convergence of AI-enhanced compression and real-time neural rendering is expected to reduce production times for immersive content by up to 70%. These developments, alongside partnerships like the Adobe-NVIDIA GTC 2026 agreement for cloud-native 3D digital twins, suggest that the primary industry challenge has shifted from rendering speed to the efficient transmission and storage of volumetric data—a gap GenSplatCodec and similar generative codecs aim to fill.
Read full article at arxiv.org
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source