Spec-Gaussian: Anisotropic View-Dependent Appearance for 3D Gaussian Splatting

Authors: Ziyi Yang, Xinyu Gao, Yang-Tian Sun, Yihua Huang, Xiaoyang Lyu, Wen Zhou, Shaohui Jiao, Xiaojuan Qi, Xiaogang Jin

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental results demonstrate that our method surpasses existing approaches in terms of rendering quality. In this section, we present both quantitative and qualitative results of our method.
Researcher Affiliation Collaboration 1State Key Lab of CAD&CG, Zhejiang University 2The University of Hong Kong 3Byte Dance Inc.
Pseudocode No The paper describes its methods but does not include structured pseudocode or algorithm blocks.
Open Source Code Yes Our codes and datasets are available at https://ingra14m.github.io/Spec-Gaussian-website.
Open Datasets Yes We used the Ne RF, NSVF, and our "Anisotropic Synthetic" datasets as the experimental datasets for synthetic scenes. Our codes and datasets are available at https://ingra14m.github.io/Spec-Gaussian-website.
Dataset Splits Yes We used the Ne RF, NSVF, and our "Anisotropic Synthetic" datasets as the experimental datasets for synthetic scenes. We used the Mip360 [2] dataset, which contains indoor scenes with specular highlights.
Hardware Specification Yes All experiments were conducted on an NVIDIA RTX 3090.
Software Dependencies No The paper states "We implemented our framework using Py Torch [42]" but does not provide a specific version number for PyTorch or other key software dependencies.
Experiment Setup Yes For the ASG appearance field, the decoupling MLP Ψ consists of 3 layers, each with 64 hidden units, and the positional encoding for the view direction is of order 2. Regarding coarse-to-fine training, which is applied only to real-world scenes to remove floaters, we start with a resolution rs that is 4x downsampled. [...] We optimize the learnable parameters and MLPs using the same loss function as 3D-GS [23]. The total supervision is given by: L = (1 − λD-SSIM)L1 + λD-SSIMLD-SSIM, where the λD-SSIM = 0.2 is consistently used in our experiments.