Compression News
The latest Compression coverage curated for game developers.
arXiv cs.GR details DELUGE, a particle-streaming compression scheme aimed at real-time VR and AR collaboration. It targets fluid, smoke, and granular sims with sub-frame latency, and the authors say it …
A new Gaussian-encoding method can compress 3D field data across structured grids, meshes, and particles under one fixed budget. The approach claims up to 4.8 dB better PSNR while using …
A new compression method for 3D Gaussian splats targets one of the format’s biggest practical costs: color attributes. By transforming spherical harmonic coefficients before coding, the researchers report more than …
A new Gaussian-volume encoding method packs scalar fields into a fixed primitive budget, then refines those primitives directly against the source data without densification or pruning. In tests, it cut …
A new visual analytics framework called FZ-VIS tackles one of scientific computing’s hardest compression problems: keeping quantities of interest intact while still shrinking huge simulation datasets. It gives users a …
A new sparse volume rendering method, ESVR, pushes large-scale volumetric visualization toward real-time use with ellipsoid primitives instead of dense voxels. It learns directly from 3D data, supports interactive transfer-function …
A new 4D Gaussian splatting format called TSOG aims to cut dynamic-scene storage by more than 90% while keeping visual quality close to existing representations. It adds temporal ordering and …
AtlasLC targets a practical bottleneck in 3D Gaussian Splatting: making compressed object assets fast enough for XR pipelines. The source-free, training-free method works directly on released Gaussian data, cutting atlas …
ECoNGS brings compressive neural Gaussian splatting to volume visualization, aiming to cut the bloat of per-scene primitive representations. The system uses lightweight networks to predict editable splats from anchor points, …
3D Gaussian Splatting streams are getting a robustness pass: a new compression pipeline packages each anchor atomically so packet loss turns into clean missing anchors instead of mangled attributes. The …
A new learned volume-compression system called EVOLVE targets high compression ratios for scientific simulation data without the per-volume tuning cost of implicit neural methods. Built around an autoencoder and variable-rate …
Lossless-INR turns volumetric neural representations into exact reconstruction instead of “close enough” compression. The method decomposes voxel values into bit-planes, then learns each bit as a binary classification problem. For …
A new INR training approach turns continuous regression into bin-based classification so lightweight neural volume encoders can predict their own error distribution. For graphics and tools teams, the appeal is …
A new image-compression method for 2D Gaussian representations cuts bitrate by about 20% while keeping visual quality on par. Cluster-Guided Vector Quantization groups similar Gaussian parameters before quantization, improving rate-distortion …
A new 3D Gaussian Splatting pruning method uses coreset theory to cut scenes down with provable guarantees instead of pure heuristics. The big practical angle is aggressive compression for deployment …
A new arXiv paper proposes shifting satellite imaging from raw-image delivery to semantic delivery: send only mission-relevant information, not every pixel. SpaceRipple uses adaptive compression on the sensing satellite, then …
The interesting bit here is not better compression in the abstract, but renderable partial states: the scene stays usable even when delivery is incomplete. The paper reports that nominal refinement …
SelfBootTok splits image tokens into global and local groups, cutting redundant detail from the generator’s job. The paper reports roughly 40% less generator computation while improving reconstruction and generation quality. …
GPU-side compression/decompression for dynamic Gaussian splats is the real story here: GS-NFS claims full-frame-rate streaming with 1-2 orders of magnitude better encode/decode speed than prior work. For teams watching 3D …
A post-training compression method for 3D Gaussian Splatting cuts model size without retraining, which is the part that should catch graphics programmers' attention. Across 13 scenes, the authors report 3.95x, …