Spatial Pruned Sparse Convolution for Efficient 3D Object Detection
Authors: Jianhui Liu, Yukang Chen, Xiaoqing Ye, Zhuotao Tian, Xiao Tan, Xiaojuan Qi
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on the KITTI, Waymo and nu Scenes datasets demonstrate that our method can achieve more than 50% reduction in GFLOPs without compromising the performance. |
| Researcher Affiliation | Collaboration | Jianhui Liu1 Yukang Chen2 Xiaoqing Ye3 Zhuotao Tian2 Xiao Tan3 Xiaojuan Qi1 1The University of Hong Kong 2The Chinese University of Hong Kong 3Baidu |
| Pseudocode | No | The paper does not contain any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | Code will be released upon acceptance. |
| Open Datasets | Yes | We evaluate our method on three challenging benchmarks KITTI [13], Waymo [26] and nu Scenes [4]. |
| Dataset Splits | Yes | KITTI dataset consists of 7,481 samples and 7,518 testing samples, where the training samples are generally divided into the train split (3,712 samples) and the val split (3,769 samples). Waymo dataset contains 1,000 sequences in total, with 798 for training and 202 for validation. nu Scenes dataset... It is split into 700 scenes for training, 150 scenes for validation, and 150 scenes for testing. |
| Hardware Specification | No | The paper mentions GFLOPs reduction, but does not specify any particular hardware (e.g., GPU/CPU models, cloud provider) used for experiments. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers. |
| Experiment Setup | Yes | Our modules contain one hyperparameter, i.e., pruning ratio and we choose to use top-k to divide the pruning part. For spatial pruned submanifold sparse convolution (SPSS-Conv) it refers to the proportion of positions in each stage that can be ignored for calculation, which can be symbolized as {rs0, rs1, rs2, rs3}. We set it as {0.3, 0.3, 0.3, 0.3 } for nu Scenes and {0.5, 0.5, 0.5, 0.5} for Waymo and KITTI. As for spatial pruned regular sparse convolution (SPRS-Conv), the pruning ratio is used for controlling the amount that needs to be inflated when downsampling. We symbolized it as {rd1, rd2, rd2}, which are set as {0.5, 0.5, 0.5} in nu Scenes and Waymo and {0.7, 0.5, 0.3} in KITTI. |