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..
VoxDet: Rethinking 3D Semantic Scene Completion as Dense Object Detection
Authors: Wuyang Li, Zhu Yu, Alexandre Alahi
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Experiments 5.1 Comparison with State-of-the-art Methods ... 5.2 Quantitative Study ... Ablation Study. Tab. 4 presents the ablation study, highlighting the following insightful observations. |
| Researcher Affiliation | Academia | Wuyang Li1 Zhu Yu2 Alexandre Alahi1 2Zhejiang University 1École Polytechnique Fédérale de Lausanne (EPFL) EMAIL EMAIL EMAIL |
| Pseudocode | Yes | C.3 Algorithmic details of the Vox NT Trick We present the detailed process of our Vox NT Trick in Algorithm 1 with Py Torch-style pseudo code. |
| Open Source Code | Yes | All data used in this study are drawn from publicly accessible platforms and employed following existing works. The cleaned codebase and trained models, including a comprehensive README, have been released publicly. |
| Open Datasets | Yes | Following the unified practice in SSC, we evaluate our approach on two benchmarks: Semantic KITTI [4] and SSCBench-KITTI-360 [33], which are derived from the original KITTI Odometry [16] and KITTI-360 [37] datasets, respectively. |
| Dataset Splits | Yes | C.1 Datasets and Metrics Benchmark Setting. Following the unified practice in SSC, we evaluate our approach on two benchmarks: Semantic KITTI [4] and SSCBench-KITTI-360 [33]... (1) Semantic KITTI consists of 22 sequences (00–21) of LiDAR scans and synchronized stereo images. We adopt the standard split of 10 sequences for training (00–07, 09–10), one sequence for validation (08), and 11 sequences for online evaluation on the hidden test server (11–21). ... (2) SSCBench-KITTI-360 is a recently released extension that re-labels a subset of the KITTI-360 sequences for SSC. It comprises 9 sequences in total, of which 7 are used for training, 1 for validation, and 1 held out for final testing. |
| Hardware Specification | Yes | Model Training. We train our Vox Det with a batch size of 4 using Adam W [42] optimizer... The experiments are conducted on 2 NVIDIA A100 GPUs (40G) with two samples for each GPU. ... The efficiency experiments are conducted on a NVIDIA 4090 (commercial GPU), considering the more practical deployment property. |
| Software Dependencies | No | Model Training. We train our Vox Det with a batch size of 4 using Adam W [42] optimizer. |
| Experiment Setup | Yes | Model Training. We train our Vox Det with a batch size of 4 using Adam W [42] optimizer. Following [79], the cosine annealing schedule is adopted, with the first 5% iterations of warm-up, maximum learning rate of 3e-4, weight decay of 0.01 and beta1 = 0.9, beta2 = 0.99, The experiments are conducted on 2 NVIDIA A100 GPUs (40G) with two samples for each GPU. |