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..
High Dynamic Range Imaging with Time-Encoding Spike Camera
Authors: Zhenkun Zhu, Ruiqin Xiong, Jiyu Xie, Yuanlin Wang, Xinfeng Zhang, Tiejun Huang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 6 Experiments 6.1 Spike Simulator 6.2 Dataset 6.3 Implementation Details 6.4 Comparison with ML Spike Camera 6.5 Ablation Studies |
| Researcher Affiliation | Collaboration | 1State Key Laboratory of Multimedia Information Processing, School of Computer Science, Peking University 2Shanghai Radio Equipment Research Institute, Shanghai, China 3School of Computer Science and Technology, University of Chinese Academy of Sciences |
| Pseudocode | No | The paper describes the working mechanism and reconstruction steps in detail (e.g., Section 4 and Section 5) and includes figures (e.g., Figure 2 and Figure 3) to illustrate architectures and mechanisms. However, it does not present any explicitly labeled pseudocode or algorithm blocks with structured steps in a code-like format. |
| Open Source Code | No | 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: [No] Justification: We will release our code after publication. |
| Open Datasets | Yes | The source codes and datasets are available at https://github.com/zkzhu123/TESC. We use part of HDM-HDR-2014 dataset [13] for training and the other part of HDM-HDR-2014 dataset, along with Kalantari13 dataset [22] for testing. |
| Dataset Splits | Yes | We use part of HDM-HDR-2014 dataset [13] for training and the other part of HDM-HDR-2014 dataset, along with Kalantari13 dataset [22] for testing. We partition HDM-HDR-2014 dataset into sub-sequences, each containing a continuous scene, with a maximum length of 400 frames and no repeated frames between sub-sequences. This results in a total of 57 sub-sequences. We utilize 30 sub-sequences for training and reserve the remaining 27 sub-sequences for testing. For the training set, we resize the original images from 1920 1080 to 240 135 to facilitate fast training. For the testing set, we crop the original images to a resolution of 512 384. |
| Hardware Specification | Yes | All tests are conducted on an Ubuntu 20.04 system with an Intel Core i7 CPU and an RTX 3090 GPU. |
| Software Dependencies | No | The paper mentions "Ubuntu 20.04 system" for the operating system, but does not specify versions for other key software components or libraries (e.g., Python, PyTorch, CUDA, etc.) that would be necessary for reproduction. |
| Experiment Setup | Yes | The network is trained for 60 epochs with a batch size of 4. We use the Adam [24] optimizer with parameter β1 = 0.9 and β2 = 0.999. The initial learning rate is set to 1e 4 and is halved every 10 epochs. To guide the training, we employ the L1 loss function to compute the difference between the estimated ˆI(t0) and the ground truth Igt(t0): L = ||ˆI(t0)/η Igt(t0)||1. |