ROI-GS: Interest-based Local Quality 3D Gaussian Splatting

2025-10-02

AI Summary

Existing 3D Gaussian Splatting (3DGS) employs uniform resource allocation, leading to suboptimal fidelity in regions of interest (ROIs) and causing model redundancy and increased rendering overhead. To address this, we propose an object-aware local refinement framework: it dynamically focuses training on ROIs via object-guided camera selection and targeted optimization; introduces high-fidelity local reconstruction coupled with multi-scale global fusion to jointly enhance geometric and appearance details while compressing the model. Our method preserves real-time rendering capability while significantly improving local reconstruction quality. Experiments on single-object ROI scenes demonstrate a 2.96 dB PSNR gain, a 17% reduction in model size, and accelerated training—achieving superior overall performance compared to state-of-the-art 3DGS approaches.

Technology Category

• Computer Vision: 3D Computer Vision
• Search and Optimization: Local Search
• Machine Learning: Optimization

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

We tackle the challenge of efficiently reconstructing 3D scenes with high detail on objects of interest. Existing 3D Gaussian Splatting (3DGS) methods allocate resources uniformly across the scene, limiting fine detail to Regions Of Interest (ROIs) and leading to inflated model size. We propose ROI-GS, an object-aware framework that enhances local details through object-guided camera selection, targeted Object training, and seamless integration of high-fidelity object of interest reconstructions into the global scene. Our method prioritizes higher resolution details on chosen objects while maintaining real-time performance. Experiments show that ROI-GS significantly improves local quality (up to 2.96 dB PSNR), while reducing overall model size by $approx 17%$ of baseline and achieving faster training for a scene with a single object of interest, outperforming existing methods.

Problem

• Enhancing local detail on 3D objects of interest
• Reducing model size in 3D Gaussian Splatting methods
• Improving reconstruction efficiency for targeted object regions

Innovation

• Object-guided camera selection for detail enhancement
• Targeted object training to prioritize regions of interest
• Seamless integration of high-fidelity objects into global scenes