GS-PQM: A Parameter-Domain Quality Metric for Compressed Gaussian Splatting
GS-PQM is a new quality metric for assessing post-training compression in Gaussian Splatting models by analyzing their internal parameters directly, avoiding costly rendering. It compares uncompressed and compressed models using a novel distortion measure, GS-Dist, and then uses an SVR model to map these parameter differences to human perception scores. This provides an accurate objective way to judge compression quality.