CAGroup3D: Class-Aware Grouping for 3D Object Detection on Point Clouds

Authors: Haiyang Wang, Lihe Ding, Shaocong Dong, Shaoshuai Shi, Aoxue Li, Jianan Li, Zhenguo Li, Liwei Wang

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our model achieves state-of-the-art 3D detection performance with remarkable gains of +3.6% on Scan Net V2 and +2.6% on SUN RGB-D in term of m AP@0.25. and We conduct extensive ablation studys on the val sets of Scan Net V2 to analyze individual components of our proposed method.
Researcher Affiliation Collaboration 1Center for Data Science, Peking University 2Beijing Institute of Technology 3Max Planck Institute for Informatics 4Huawei Noah s Ark Lab, China 5National Key Laboratory of General Artificial Intelligence, School of Intelligence Science and Technology, Peking University
Pseudocode No The paper describes methods in prose and figures, but does not contain structured pseudocode or algorithm blocks.
Open Source Code No Code will be available at https://github.com/Haiyang-W/CAGroup3D.
Open Datasets Yes Our CAGroup3D is evaluated on two indoor challenging 3D scene datasets, i.e., Scan Net V2 [6] and SUN RGB-D [33].
Dataset Splits Yes For all datasets, we follow the standard data splits adopted in [27]. Scan Net V2 contains richly-annotated 3D reconstructed indoor scenes with axis-aligned bounding box for most common 18 object categories. It contains 1201 training samples and the remaining 312 scans are left for validation.
Hardware Specification Yes All models are trained on two NVIDIA Tesla V100 GPUs with a 32 GB memory per-card.
Software Dependencies No The paper mentions 'Adam W optimizer [23]' and 'Mind Spore*' but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes For both datasets, we set the voxel size as 0.02m. ... Training Scan Net V2 requires 120 epochs with the learning rate decay by 10x on 80 epochs and 110 epochs. SUN RGB-D takes 48 epochs and learning rate decayed on 32, 44 epochs. ... we set batch size, initial learning rate and weight decay are 16, 0.001 and 0.0001 for both datasets.