{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/latte-accelerating-lidar-point-cloud","title":"LATTE: Accelerating LiDAR Point Cloud Annotation via Sensor Fusion, One-Click Annotation, and Tracking","arxiv_id":"1904.09085","date":"2019-04-19","proceeding":null,"authors":["Bernie Wang","Virginia Wu","Bichen Wu","Kurt Keutzer"],"abstract":"LiDAR (Light Detection And Ranging) is an essential and widely adopted sensor\nfor autonomous vehicles, particularly for those vehicles operating at higher\nlevels (L4-L5) of autonomy. Recent work has demonstrated the promise of\ndeep-learning approaches for LiDAR-based detection. However, deep-learning\nalgorithms are extremely data hungry, requiring large amounts of labeled\npoint-cloud data for training and evaluation. Annotating LiDAR point cloud data\nis challenging due to the following issues: 1) A LiDAR point cloud is usually\nsparse and has low resolution, making it difficult for human annotators to\nrecognize objects. 2) Compared to annotation on 2D images, the operation of\ndrawing 3D bounding boxes or even point-wise labels on LiDAR point clouds is\nmore complex and time-consuming. 3) LiDAR data are usually collected in\nsequences, so consecutive frames are highly correlated, leading to repeated\nannotations. To tackle these challenges, we propose LATTE, an open-sourced\nannotation tool for LiDAR point clouds. LATTE features the following\ninnovations: 1) Sensor fusion: We utilize image-based detection algorithms to\nautomatically pre-label a calibrated image, and transfer the labels to the\npoint cloud. 2) One-click annotation: Instead of drawing 3D bounding boxes or\npoint-wise labels, we simplify the annotation to just one click on the target\nobject, and automatically generate the bounding box for the target. 3)\nTracking: we integrate tracking into sequence annotation such that we can\ntransfer labels from one frame to subsequent ones and therefore significantly\nreduce repeated labeling. Experiments show the proposed features accelerate the\nannotation speed by 6.2x and significantly improve label quality with 23.6% and\n2.2% higher instance-level precision and recall, and 2.0% higher bounding box\nIoU. LATTE is open-sourced at https://github.com/bernwang/latte.","url_abs":"http://arxiv.org/abs/1904.09085v1","url_pdf":"http://arxiv.org/pdf/1904.09085v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"latte-accelerating-lidar-point-cloud","repo_url":"https://github.com/bernwang/latte","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"latte-accelerating-lidar-point-cloud","repo_url":"https://github.com/Rabeea369/latte","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"autonomous-vehicles","task_name":"Autonomous Vehicles"},{"task_slug":"sensor-fusion","task_name":"Sensor Fusion"}],"methods":[{"method_slug":"speed","method_name":"SPEED"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1904.09085","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}