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..
Event-Driven Dynamic Scene Depth Completion
Authors: Zhiqiang Yan, Jianhao Jiao, Zhengxue Wang, Gim Hee Lee
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on this benchmark demonstrate the superiority of our Event DC. |
| Researcher Affiliation | Academia | 1National University of Singapore 2University College London 3Nanjing University of Science and Technology |
| Pseudocode | No | The paper describes its methodology in Section 3, 'Our Method', using mathematical formulations and descriptive text, but does not present any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The data and source codes will be publicly available upon acceptance. |
| Open Datasets | Yes | Event DC-Semi Syn is a semi-synthetic dataset based on KITTI [14]. |
| Dataset Splits | Yes | Table 1: Basic statistics of the Event DC benchmark. Dataset Color Camera Depth Sensor Event Camera Train Test Resolution Event DC-Real FLIR BFS-U3-31S4C Ouster OS1-128 Li DAR DAVIS346 14,845 1,000 320 256 Event DC-Semi Syn Point Grey Flea2 Velodyne HDL-64E Li DAR 7,094 2,213 1216 256 Event DC-Full Syn 21,000 500 512 256 |
| Hardware Specification | Yes | We implement Event DC using the Py Torch framework and conduct training on two NVIDIA RTX 4090 GPUs using the Distributed Data Parallel strategy for efficiency. |
| Software Dependencies | No | The paper mentions 'Py Torch framework' but does not specify a version number or provide version numbers for other software dependencies like AdamW or One Cycle learning rate policy. |
| Experiment Setup | Yes | Optimization is performed with the Adam W optimizer [31] in conjunction with the One Cycle learning rate policy [43]. The training process begins with a warm-up stage that linearly increases the learning rate from 0.00002 to 0.001 over the first 10% of iterations. Subsequently, a cosine annealing schedule gradually decays the learning rate to a final value of 0.0002. The batch size is set to 2 per GPU. In addition, to further enhance model performance, we employ a set of data augmentation strategies [46, 29], including random horizontal flip, rotation, cropping, and color jitter. [...] where ฮป and ยต are weighting hyper-parameters that we empirically set to 1 and 0.1, respectively. |