Reproduce “3D Gaussian Splatting for Fine-grained Urban Scene Reconstruction via Semantic-guided Density Control and Loss Design” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yueling Liao, San Jiang, Xing Zhang, Zhongwen Hu
SG-GS uses segmentation from SAM to direct 3D Gaussian placement and training supervision toward structures and boundaries. It classifies image regions to adapt densification thresholds, then weights pixel, edge, and frequency losses during training. On Derrick, held-out-view PSNR rises from 25.84 dB for 3DGS to 27.18 dB; gains are not universal across metrics or methods, and training takes 40% longer.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.