{"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/pre-training-by-completing-point-clouds","title":"Unsupervised Point Cloud Pre-Training via Occlusion Completion","arxiv_id":"2010.01089","date":"2020-10-02","proceeding":"ICCV 2021 10","authors":["Hanchen Wang","Qi Liu","Xiangyu Yue","Joan Lasenby","Matthew J. Kusner"],"abstract":"We describe a simple pre-training approach for point clouds. It works in three steps: 1. Mask all points occluded in a camera view; 2. Learn an encoder-decoder model to reconstruct the occluded points; 3. Use the encoder weights as initialisation for downstream point cloud tasks. We find that even when we construct a single pre-training dataset (from ModelNet40), this pre-training method improves accuracy across different datasets and encoders, on a wide range of downstream tasks. Specifically, we show that our method outperforms previous pre-training methods in object classification, and both part-based and semantic segmentation tasks. We study the pre-trained features and find that they lead to wide downstream minima, have high transformation invariance, and have activations that are highly correlated with part labels. Code and data are available at: https://github.com/hansen7/OcCo","url_abs":"https://arxiv.org/abs/2010.01089v3","url_pdf":"https://arxiv.org/pdf/2010.01089v3.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":"pre-training-by-completing-point-clouds","repo_url":"https://github.com/hansen7/OcCo","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[{"task_slug":"3d-point-cloud-linear-classification","task_name":"3D Point Cloud Linear Classification"},{"task_slug":"decoder","task_name":"Decoder"},{"task_slug":"few-shot-3d-point-cloud-classification","task_name":"Few-Shot 3D Point Cloud Classification"},{"task_slug":"point-cloud-classification","task_name":"Point Cloud Classification"},{"task_slug":"point-cloud-pre-training","task_name":"Point Cloud Pre-training"},{"task_slug":"point-cloud-segmentation","task_name":"Point Cloud Segmentation"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"},{"task_slug":"unsupervised-3d-point-cloud-linear-evaluation","task_name":"Unsupervised 3D Point Cloud Linear Evaluation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/3d-point-cloud-linear-classification-on","task":"3D Point Cloud Linear Classification","dataset":"ModelNet40","model":"OcCo","rank_in_archive_order":16,"of":20,"metrics":{"Overall Accuracy":"89.2"},"uses_additional_data":false},{"leaderboard":"/sota/3d-point-cloud-linear-classification-on-1","task":"3D Point Cloud Linear Classification","dataset":"ScanObjectNN","model":"OcCo","rank_in_archive_order":4,"of":4,"metrics":{"Overall Accuracy":"78.3"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-3d-point-cloud-classification-on-3","task":"Few-Shot 3D Point Cloud Classification","dataset":"ModelNet40 10-way (10-shot)","model":"OcCo+PointNet","rank_in_archive_order":20,"of":31,"metrics":{"Overall Accuracy":"83.9","Standard Deviation":"1.8"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-3d-point-cloud-classification-on-3","task":"Few-Shot 3D Point Cloud Classification","dataset":"ModelNet40 10-way (10-shot)","model":"OcCo+DGCNN","rank_in_archive_order":21,"of":31,"metrics":{"Overall Accuracy":"82.9","Standard Deviation":"1.3"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-3d-point-cloud-classification-on-4","task":"Few-Shot 3D Point Cloud Classification","dataset":"ModelNet40 10-way (20-shot)","model":"OcCo+PointNet","rank_in_archive_order":20,"of":31,"metrics":{"Overall Accuracy":"89.7","Standard Deviation":"1.5"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-3d-point-cloud-classification-on-4","task":"Few-Shot 3D Point Cloud Classification","dataset":"ModelNet40 10-way (20-shot)","model":"OcCo+DGCNN","rank_in_archive_order":21,"of":31,"metrics":{"Overall Accuracy":"86.5","Standard Deviation":"2.2"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-3d-point-cloud-classification-on-1","task":"Few-Shot 3D Point Cloud Classification","dataset":"ModelNet40 5-way (10-shot)","model":"OcCo+DGCNN","rank_in_archive_order":19,"of":30,"metrics":{"Overall Accuracy":"90.6","Standard Deviation":"2.8"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-3d-point-cloud-classification-on-1","task":"Few-Shot 3D Point Cloud Classification","dataset":"ModelNet40 5-way (10-shot)","model":"OcCo+PointNet","rank_in_archive_order":20,"of":30,"metrics":{"Overall Accuracy":"89.7","Standard Deviation":"1.9"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-3d-point-cloud-classification-on-2","task":"Few-Shot 3D Point Cloud Classification","dataset":"ModelNet40 5-way (20-shot)","model":"OcCo+DGCNN","rank_in_archive_order":19,"of":30,"metrics":{"Overall Accuracy":"92.5","Standard Deviation":"1.9"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-3d-point-cloud-classification-on-2","task":"Few-Shot 3D Point Cloud Classification","dataset":"ModelNet40 5-way (20-shot)","model":"OcCo+PointNet","rank_in_archive_order":20,"of":30,"metrics":{"Overall Accuracy":"92.4","Standard Deviation":"1.6"},"uses_additional_data":false},{"leaderboard":"/sota/few-shot-3d-point-cloud-classification-on-6","task":"Few-Shot 3D Point Cloud Classification","dataset":"ScanObjectNN 10-way (10-shot)","model":"OcCo","rank_in_archive_order":3,"of":3,"metrics":{"Overall Accuracy":"57.0"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-classification-on-pointcloud-c","task":"Point Cloud Classification","dataset":"PointCloud-C","model":"OcCo-DGCNN","rank_in_archive_order":20,"of":24,"metrics":{"mean Corruption Error (mCE)":"1.047"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-segmentation-on-pointcloud-c","task":"Point Cloud Segmentation","dataset":"PointCloud-C","model":"OcCo-DGCNN","rank_in_archive_order":5,"of":11,"metrics":{"mean Corruption Error (mCE)":"0.977"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-segmentation-on-pointcloud-c","task":"Point Cloud Segmentation","dataset":"PointCloud-C","model":"OcCo-PointNet","rank_in_archive_order":9,"of":11,"metrics":{"mean Corruption Error (mCE)":"1.130"},"uses_additional_data":false},{"leaderboard":"/sota/point-cloud-segmentation-on-pointcloud-c","task":"Point Cloud Segmentation","dataset":"PointCloud-C","model":"OcCo-PCN","rank_in_archive_order":10,"of":11,"metrics":{"mean Corruption Error (mCE)":"1.173"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2010.01089","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}