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..
Spik-NeRF: Spiking Neural Networks for Neural Radiance Fields
Authors: Gang Wan, Qinlong Lan, Zihan Li, Huimin Wang, Wu Yitian, wang zhen, Wanhua Li, Yufei Guo
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate Spik-Ne RF on both realistic and synthetic scenes, and the experimental results demonstrate that Spik-Ne RF achieves rendering performance comparable to ANN-based Ne RF models. We evaluate the rendering performance of Spik-Ne RF both quantitatively and qualitatively. Tables 1 and 2 present per-scene quantitative results from the synthetic and realistic datasets, respectively. |
| Researcher Affiliation | Collaboration | Gang Wan1, , Qinlong Lan1, , Zihan Li2,3, Huimin Wang2,3, Yitian Wu1, Zhen Wang1, Wanhua Li1, Yufei Guo3, 1Space Engineering University, 2Peking University 3Intelligent Science & Technology Academy of CASIC |
| Pseudocode | No | The paper describes methods through mathematical equations and textual descriptions (e.g., equations 1-28), but does not contain a structured pseudocode or algorithm block. |
| Open Source Code | Yes | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] . Justification: We provide open access to the data and code with sufficient instructions in the supplemental material. |
| Open Datasets | Yes | We assess the rendering performance of Spik-Ne RF on both synthetic and real-world datasets [31]. The synthetic dataset includes eight scenes featuring different objects. For each scene, there are 100 views used for training and 200 views for testing, with each view image having a resolution of 400 x 400 pixels. The real-world dataset consists of eight scenes captured with mobile phones. |
| Dataset Splits | Yes | For each scene, there are 100 views used for training and 200 views for testing, with each view image having a resolution of 400 x 400 pixels. The real-world dataset consists of eight scenes captured with mobile phones. Each scene contains between 20 and 60 images, and the images are resized to 400 x 400 pixels in this paper. Additionally, one-eighth of the images are reserved for testing. |
| Hardware Specification | No | The paper states that 'The computation resources description is provided in the experiment section' in the NeurIPS checklist, but the experiment section only describes software (Adam optimizer) and training parameters (iterations, batch size), not specific hardware details like GPU or CPU models. |
| Software Dependencies | No | The paper mentions 'Adam optimizer' as part of the training process but does not provide specific version numbers for any software dependencies, libraries, or programming languages used. |
| Experiment Setup | Yes | All models are trained using the Adam optimizer for 300,000 iterations with a batch size of 1,024 rays. We initialize the learning rate at 5e-4, which is decayed exponentially as training progresses. For synthetic scenes, the number of sampled points is set to 64 for the coarse network and 128 for the fine network. Similarly, for real-world scenes, 64 and 128 sampled points are used for the coarse and fine networks, respectively. The total number of timesteps for both Spiking Ne RF and our Spik-Ne RF is set to 2, while for the Spiking-Ne RF, it is 8 timesteps. |