This repository provides a pipeline for 3D reconstruction and relighting using 3D Gaussian Splatting from a single-view image. The implementation is based on TriplaneGaussian and Relightable 3D Gaussian. It leverages PyTorch and related deep-learning frameworks to achieve high-quality relighting and reconstruction.

Follow the installation guides for each module:

• Triplane Gaussian Installation: For single-view 3D reconstruction using Triplane Gaussian.
• Relightable 3D Gaussian Installation: For real-time relighting using BRDF decomposition and ray tracing.
• Colmap to NeRF Transformation: For generating transform.jason file from the video output of Triplane Gaussian.

python infer.py --config config.yaml data.image_list=[example_images/a_pikachu_with_smily_face.webp] --image_preprocess

python ----

• TriplaneGaussian Repository
• Relightable 3D Gaussian Repository
• Facebook Research - PyTorch3D
• GraphDeco Inria - Diff Gaussian Rasterization