Compressible-composable NeRF via Rank-residual Decomposition

Authors: Jiaxiang Tang, Xiaokang Chen, Jingbo Wang, Gang Zeng

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate that our method is able to achieve comparable rendering quality to state-of-the-art methods, while enabling extra capability of compression and composition. Code is available at https://github.com/ashawkey/CCNe RF.
Researcher Affiliation Academia Jiaxiang Tang1, Xiaokang Chen1, Jingbo Wang2, Gang Zeng1,3 1School of Intelligence Science and Technology, Peking University 2Chinese University of Hong Kong 3Intelligent Terminal Key Laboratory of Si Chuan Province {tjx, pkucxk}@pku.edu.cn, wj020@ie.cuhk.edu.hk, zeng@pku.edu.cn
Pseudocode No The paper describes its methodology in narrative form with mathematical equations, but does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes Code is available at https://github.com/ashawkey/CCNe RF.
Open Datasets Yes We mainly carry out experiments on the Ne RF-synthetic dataset [26] (CC BY 3.0 license) and the Tanks and Temples dataset [17] (CC BY-NC-SA 3.0 license).
Dataset Splits No The paper mentions using the NeRF-synthetic and Tanks and Temples datasets, but does not explicitly provide specific percentages, sample counts, or detailed methodology for training, validation, and test splits.
Hardware Specification Yes All the experiments are performed on one NVIDIA V100 GPU.
Software Dependencies No The model is implemented with the Py Torch framework [32]. However, a specific version number for PyTorch or other software dependencies is not provided.
Experiment Setup Yes We use the Adam optimizer [16] with an initial learning rate of 0.02 for the factorized matrices, and 0.001 for the singular values.