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..
Multimodal LiDAR-Camera Novel View Synthesis with Unified Pose-free Neural Fields
Authors: Weiyi Xue, Fan Lu, Yunwei Zhu, Zehan Zheng, Sanqing Qu, Jiangtong Li, Ya Wu, Haiyun Wei, Guang Chen
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our method across diverse scenarios using the KITTI-360 [15] and Nu Scenes [4] autonomous driving datasets. Comprehensive experiments demonstrate that MUP significantly outperforms prior state-of-the-art techniques and single-modality approaches by a large margin in both registration and NVS. |
| Researcher Affiliation | Collaboration | 1 Tongji University, 2 Shanghai Innovation Institute, 3 CNNC Equipment Technology Research (Shanghai) Co., Ltd. |
| Pseudocode | No | The paper describes the methodology in Section 4 using prose and figures (e.g., Figure 2 and 3) but does not present any formal pseudocode or algorithm blocks. |
| 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: This paper provide open access to the data and code. |
| Open Datasets | Yes | We evaluate our method across diverse scenarios using the KITTI-360 [15] and Nu Scenes [4] autonomous driving datasets. |
| Dataset Splits | No | Following [47] we selected 33 consecutive frames from keyframes as a single scene. KITTI-360 has three cameras and a Li DAR, where each frame s point cloud and image are time-aligned. Following [36, 55, 37], we use the standard KITTI-360 dataset, all images and point clouds are time-synchronized. For both datasets, only the front-facing single camera was utilized. |
| Hardware Specification | Yes | All experiments were conducted on a single NVIDIA Ge Force RTX 3090 GPU. |
| Software Dependencies | No | MUP optimization was implemented in Py Torch [26] using the Adam optimizer [13]. |
| Experiment Setup | Yes | The learning rates were set as follows: 1 10 2, decaying to 1 10 4 for Ne RF; 1 10 3, decaying to 1 10 5 for translation; and 5 10 3, decaying to 5 10 5 for rotation. The weighting coefficients for each loss term are defined as: λα = λβ = 1000, λγ = 10, λη = 2.5, λr = 150. Besides, all sequences are trained for 60K iterations in the pose-free setting and 30K iterations when ground truth poses are available. |