GScomp-QA dataset benchmarks compressed Gaussian Splatting quality

Gaussian Splatting (GS) has become a leading method for high-quality 3D reconstruction and novel view synthesis, but its large model size poses storage and transmission challenges. While multiple GS compression solutions have been proposed, their perceptual impact on viewers remains poorly understood due to the lack of dedicated evaluation datasets. To fill this gap, a team of researchers — Pedro Martin, António Rodrigues, João Ascenso, and Maria Paula Queluz — introduces GScomp-QA, a subjective quality assessment dataset tailored specifically for compressed GS models.

The dataset comprises 331 video stimuli derived from 13 real-world scenes, representing 9 state-of-the-art GS compression methods. By using videos synthesized from uncompressed models as reference, GScomp-QA successfully isolates compression-induced distortions from synthesis artifacts. A subjective study with 20 participants yielded reliable perceptual scores, enabling a perceptual rate-distortion analysis of the compression solutions. Additionally, 18 objective quality metrics were evaluated, revealing that none fully capture the unique distortions introduced by GS compression. The dataset will be publicly available, providing a much-needed benchmark for developing better quality metrics and compression algorithms.