Neural Rendering
Neural rendering aims to synthesize realistic images of 3D scenes from various input data, such as multiple photographs or sparse sensor readings, achieving photorealistic novel view synthesis. Current research heavily focuses on improving the robustness and generalization capabilities of neural radiance fields (NeRFs) and related architectures like Gaussian splatting, addressing challenges like sparse data, dynamic scenes, and real-world imperfections (e.g., motion blur, low light). These advancements have significant implications for applications ranging from augmented and virtual reality to medical imaging and autonomous driving, enabling more accurate and efficient 3D scene representation and manipulation.
Papers
Deformable Model-Driven Neural Rendering for High-Fidelity 3D Reconstruction of Human Heads Under Low-View Settings
Baixin Xu, Jiarui Zhang, Kwan-Yee Lin, Chen Qian, Ying He
GM-NeRF: Learning Generalizable Model-based Neural Radiance Fields from Multi-view Images
Jianchuan Chen, Wentao Yi, Liqian Ma, Xu Jia, Huchuan Lu