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..
QuadricFormer: Scene as Superquadrics for 3D Semantic Occupancy Prediction
Authors: Sicheng Zuo, Wenzhao Zheng, Xiaoyong Han, Longchao Yang, yong pan, Jiwen Lu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on the nu Scenes and KITTI-360 datasets demonstrate that Quadric Former achieves state-of-the-art performance while maintaining superior efficiency. The paper includes a dedicated '4 Experiments' section with 'Datasets and Metrics', 'Implementation Details', 'Main Results', 'Ablation Study', and 'Visualizations'. |
| Researcher Affiliation | Collaboration | Sicheng Zuo1, Wenzhao Zheng1, , Xiaoyong Han1, Longchao Yang2 Yong Pan2 Jiwen Lu1,3 1Tsinghua University 2Li Auto Inc 3Beijing National Research Center for Information Science and Technology. The affiliations include Tsinghua University (academic) and Li Auto Inc (industry), indicating a collaborative effort. |
| Pseudocode | No | The paper describes the proposed approach and methodology in prose and mathematical formulations within Section 3 'Proposed Approach' but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available at https://github.com/zuosc19/Quadric Former. |
| Open Datasets | Yes | Extensive experiments on the nu Scenes and KITTI-360 datasets demonstrate that Quadric Former achieves state-of-the-art performance while maintaining superior efficiency. nu Scenes [3] comprises 1,000 urban driving sequences... KITTI-360 [27] comprises over 320k images... |
| Dataset Splits | Yes | nu Scenes [3]... The dataset is officially split into 700 sequences for training, 150 for validation, and 150 for testing. KITTI-360 [27]... The official split contains 7 sequences for training, 1 for validation, and 1 for testing, corresponding to 8487, 1812, and 2566 key frames, respectively. |
| Hardware Specification | Yes | The latency and memory are tested on an NVIDIA 4090 GPU with batch size one during inference, in accordance with Gaussian-based methods [18, 15]. |
| Software Dependencies | No | The paper mentions using ResNet101-DCN [13] with FCOS3D checkpoint [42] and ResNet50 [13], and AdamW for optimization, but does not specify the versions of key software components or libraries (e.g., PyTorch, TensorFlow, specific library versions). |
| Experiment Setup | Yes | The input images are at resolutions of 900 1600 for nu Scenes and 376 1408 for KITTI-360 [27] with random flipping and photometric distortion augmentations. We train our model for 20 epochs on nu Scenes and KITTI-360 with a batch of 8. For optimization, we train our model using Adam W with weight decay of 0.01, and maximum learning rate of 4 10 4, which decays with a cosine schedule. The numbers of Superquarics are set to 1600 in our main results for nu Scenes and KITTI-360. |