M. Vemparala, Anmol Singh, Ahmed Mzid, Nael Fasfous, Alexander Frickenstein, Florian Mirus, Hans-Joerg Voegel, N. Nagaraja, W. Stechele
{"title":"Pruning CNNs for LiDAR-based Perception in Resource Constrained Environments","authors":"M. Vemparala, Anmol Singh, Ahmed Mzid, Nael Fasfous, Alexander Frickenstein, Florian Mirus, Hans-Joerg Voegel, N. Nagaraja, W. Stechele","doi":"10.1109/ivworkshops54471.2021.9669256","DOIUrl":null,"url":null,"abstract":"Deep neural networks provide high accuracy for perception. However they require high computational power. In particular, LiDAR-based object detection delivers good accuracy and real-time performance, but demands high computation due to expensive feature-extraction from point cloud data in the encoder and backbone networks. We investigate the model complexity versus accuracy trade-off using reinforcement learning based pruning for PointPillars, a recent LiDAR-based 3D object detection network. We evaluate the model on the validation dataset of KITTI (80/20-splits) according to the mean average precision (mAP) for the car class. We prune the original PointPillars model (mAP 89.84) and achieve 65.8% reduction in floating point operations (FLOPs) for a marginal accuracy loss. The compression corresponds to 31.7% reduction in inference time and 35% reduction in GPU memory on GTX 1080 Ti.","PeriodicalId":256905,"journal":{"name":"2021 IEEE Intelligent Vehicles Symposium Workshops (IV Workshops)","volume":"61 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2021-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"3","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2021 IEEE Intelligent Vehicles Symposium Workshops (IV Workshops)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ivworkshops54471.2021.9669256","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 3
Abstract
Deep neural networks provide high accuracy for perception. However they require high computational power. In particular, LiDAR-based object detection delivers good accuracy and real-time performance, but demands high computation due to expensive feature-extraction from point cloud data in the encoder and backbone networks. We investigate the model complexity versus accuracy trade-off using reinforcement learning based pruning for PointPillars, a recent LiDAR-based 3D object detection network. We evaluate the model on the validation dataset of KITTI (80/20-splits) according to the mean average precision (mAP) for the car class. We prune the original PointPillars model (mAP 89.84) and achieve 65.8% reduction in floating point operations (FLOPs) for a marginal accuracy loss. The compression corresponds to 31.7% reduction in inference time and 35% reduction in GPU memory on GTX 1080 Ti.