${C}^{3}$-GS: Learning Context-aware, Cross-dimension, Cross-scale Feature for Generalizable Gaussian Splatting

2025-08-28

AI Summary

Existing methods struggle to learn discriminative and multi-view-consistent features under sparse-view conditions, leading to poor generalization and inaccurate geometry reconstruction in Gaussian Splatting (GS). To address this, we propose C³-GS—a context-aware, cross-dimensional, and cross-scale lightweight feature learning framework that jointly models pixel-level Gaussian parameters without additional supervision. By enforcing geometric consistency and enhancing rendering photorealism, C³-GS enables high-fidelity novel view synthesis without per-scene optimization. Evaluated on standard benchmarks, C³-GS achieves state-of-the-art performance in both rendering quality and cross-scene generalization, outperforming prior approaches significantly. Moreover, it strikes an effective balance between computational efficiency and reconstruction fidelity, demonstrating robustness across diverse scenes with minimal views.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Multi-instance/Multi-view Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Generalizable Gaussian Splatting aims to synthesize novel views for unseen scenes without per-scene optimization. In particular, recent advancements utilize feed-forward networks to predict per-pixel Gaussian parameters, enabling high-quality synthesis from sparse input views. However, existing approaches fall short in encoding discriminative, multi-view consistent features for Gaussian predictions, which struggle to construct accurate geometry with sparse views. To address this, we propose $mathbf{C}^{3}$-GS, a framework that enhances feature learning by incorporating context-aware, cross-dimension, and cross-scale constraints. Our architecture integrates three lightweight modules into a unified rendering pipeline, improving feature fusion and enabling photorealistic synthesis without requiring additional supervision. Extensive experiments on benchmark datasets validate that $mathbf{C}^{3}$-GS achieves state-of-the-art rendering quality and generalization ability. Code is available at: https://github.com/YuhsiHu/C3-GS.

Problem

• Encode discriminative multi-view consistent features
• Construct accurate geometry from sparse input views
• Improve feature fusion for photorealistic novel view synthesis

Innovation

• Context-aware cross-dimension cross-scale feature learning
• Three lightweight modules integrated rendering pipeline
• Enhanced feature fusion without additional supervision