Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

SAP: Exact Sorting in Splatting via Screen-Aligned Primitives

Authors: Zhanke Wang, Zhiyan Wang, Kaiqiang Xiong, Wu Jiahao, Yang Deng, Ronggang Wang

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Section 5: Experiments. Section 5.1: Experimental Setup, Dataset and Metrics, Implementation Details. Section 5.2: Results and Comparisons, Quantitative results, Qualitative results. Section 5.3: Ablation Studies. These sections clearly indicate the paper conducts empirical studies with data analysis.
Researcher Affiliation Academia All authors are affiliated with 'Guangdong Provincial Key Laboratory of Ultra High Definition Immersive Media Technology, Shenzhen Graduate School, Peking University' and 'Peng Cheng Laboratory'. Email domains include 'stu.pku.edu.cn' and 'pkusz.edu.cn', which are academic institutions.
Pseudocode No The paper describes methods and algorithms in narrative text and mathematical formulas, but it does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No In the NeurIPS Paper Checklist under section '5. Open access to data and code', the answer is '[No]' with the justification: 'We will provide our CUDA code, which is our SAP and the core component of our method. With only minor modifications, it can be integrated into existing frameworks. We also commit to releasing the full code after the paper is accepted.'
Open Datasets Yes Section 5.1 'Dataset and Metrics' states: 'We evaluated Screen-Aligned Primitives (SAP) on a diverse set of real-world datasets... Specifically, we tested our approach on various public datasets, including 9 scenes from Mip Ne RF360 (3), 2 scenes from Tanks&Temples (54), 2 scenes from Deep Blending (55).'
Dataset Splits Yes Section 5.1 'Experimental Setup' states: 'We adopt the experimental settings and dataset partitioning as established in Scaffold-GS (9).'. Appendix B.2 'Experimental Setup on the Mip-Ne RF360 Dataset' further details: 'In this paper, we follow the experimental setup of Scaffold-GS.'
Hardware Specification Yes Section 5.1 'Implementation Details' states: 'We conducted all experiments on a single NVIDIA L40S GPU.'
Software Dependencies No Section 5.1 'Implementation Details' mentions 'Our PyTorch implementation is built upon the Scaffold-GS (9) framework,' but does not provide specific version numbers for PyTorch, CUDA, or other key software dependencies.
Experiment Setup Yes Appendix B.2 'Hyperparameter Settings' details: 'Most of our hyperparameter settings follow those of Scaffold-GS. Specifically, we used 10 neural Gaussians for each anchor; We set the threshold for the average gradient magnitude (τ1) to 0.0002, and the threshold for the maximum gradient magnitude (τ2) to 0.0015; The initial learning rate of our 3D-consistent decoder is set at 0.004 and gradually annealed to 0.00004 during training; the learning rate for the features is set to 0.0025. All other hyperparameter settings follow those of Scaffold-GS to ensure a fair comparison.'