3D Gaussian Splatting Compression with Object Scalability

We introduce a framework toward scalable, finer-grained object-level 3DGS compression. First, we propose a post-training method namedRecastGS to reorganize pretrained 3DGS into a layered representationand progressively distill cumulative submodels to improve rate–distortionefficiency. Leveraging multi-view SAM predictions from user click prompts,Gaussians are further partitioned into user-defined regions of interest(ROI), enabling region-adaptive quality control without retraining. Sec-ond, built upon this reorganized region-aware layered hierarchy, a feed-forward 3DGS compression method named LayeredCGS is proposed tocompress position using a lightweight point cloud codec and attributeswith a layer-wise context model to exploit cross-layer correlations. Ex-tensive experiments show that LayeredCGS achieves 35% BD-Rate gainover the existing feed-forward method FCGS. With progressive distil-lation in RecastGS enabled, our method further outperforms most per-scene optimization methods. Moreover, the proposed method supportsROI-aware compression and flexible bitstream truncation, achieving upto 2 dB higher ROI PSNR at comparable bitrates compared with theuniform quality allocation baseline while enabling low-latency previewand progressive quality refinement. The code will be released at https://github.com/RuixiangXue/ScalableGSC.