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..
RayFusion: Ray Fusion Enhanced Collaborative Visual Perception
Authors: Shaohong Wang, Lu Bin, Xinyu Xiao, Hanzhi Zhong, Bowen Pang, Tong Wang, Zhiyu Xiang, Hangguan Shan, Eryun Liu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To evaluate Ray Fusion, we conduct extensive experiments on a real-world dataset, DAIR-V2X [38], and two simulated datasets, V2XSet [35] and OPV2V [36]. The experimental results demonstrate that Ray Fusion significantly outperforms previous works in terms of both performance and robustness across multiple datasets. |
| Researcher Affiliation | Academia | 1Zhejiang University 2Institute of Automation of Chinese Academy of Sciences |
| Pseudocode | No | The paper describes the methodology and architecture through text and figures (e.g., Figure 2), but no explicit pseudocode or algorithm blocks are present. |
| Open Source Code | Yes | The code is available at https://github.com/wangsh0111/Ray Fusion. |
| Open Datasets | Yes | We evaluate our proposed method on three multi-agent datasets: DAIR-V2X [38], V2XSet [35], and OPV2V [36]. |
| Dataset Splits | No | The paper mentions using DAIR-V2X, V2XSet, and OPV2V datasets but does not explicitly provide details about training, validation, or test dataset splits, such as percentages or sample counts used for these experiments. |
| Hardware Specification | Yes | FPS is measured on a single Ge Force RTX 3090 using the Py Torch fp32 backend. |
| Software Dependencies | No | The paper mentions 'Py Torch fp32 backend' but does not specify a version number for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | We implement Ray Fusion following Section 3 and train it for 72 epochs using the Adam W [20] optimizer. The initial learning rate is set to 1 10 4 and follows a cosine annealing decay schedule. The instance feature dimension C is set to 256, and the number of discrete depth bins D is set to 80, while the number of ego instances N and shared instances M are set to 600 and 200, respectively. We utilize three pyramid windows with receptive field thresholds of 4m, 8m, and 16m. |