{"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/take-a-photo-3d-to-2d-generative-pre-training","title":"Take-A-Photo: 3D-to-2D Generative Pre-training of Point Cloud Models","arxiv_id":"2307.14971","date":"2023-07-27","proceeding":"ICCV 2023 1","authors":["Ziyi Wang","Xumin Yu","Yongming Rao","Jie zhou","Jiwen Lu"],"abstract":"With the overwhelming trend of mask image modeling led by MAE, generative pre-training has shown a remarkable potential to boost the performance of fundamental models in 2D vision. However, in 3D vision, the over-reliance on Transformer-based backbones and the unordered nature of point clouds have restricted the further development of generative pre-training. In this paper, we propose a novel 3D-to-2D generative pre-training method that is adaptable to any point cloud model. We propose to generate view images from different instructed poses via the cross-attention mechanism as the pre-training scheme. Generating view images has more precise supervision than its point cloud counterpart, thus assisting 3D backbones to have a finer comprehension of the geometrical structure and stereoscopic relations of the point cloud. Experimental results have proved the superiority of our proposed 3D-to-2D generative pre-training over previous pre-training methods. Our method is also effective in boosting the performance of architecture-oriented approaches, achieving state-of-the-art performance when fine-tuning on ScanObjectNN classification and ShapeNetPart segmentation tasks. Code is available at https://github.com/wangzy22/TAP.","url_abs":"https://arxiv.org/abs/2307.14971v2","url_pdf":"https://arxiv.org/pdf/2307.14971v2.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":"take-a-photo-3d-to-2d-generative-pre-training","repo_url":"https://github.com/wangzy22/tap","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"3d-part-segmentation","task_name":"3D Part Segmentation"},{"task_slug":"3d-point-cloud-classification","task_name":"3D Point Cloud Classification"}],"methods":[{"method_slug":"mae","method_name":"MAE"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/3d-part-segmentation-on-shapenet-part","task":"3D Part Segmentation","dataset":"ShapeNet-Part","model":"PointMLP+TAP","rank_in_archive_order":9,"of":67,"metrics":{"Class Average IoU":"85.2","Instance Average IoU":"86.9"},"uses_additional_data":false},{"leaderboard":"/sota/3d-point-cloud-classification-on-scanobjectnn","task":"3D Point Cloud Classification","dataset":"ScanObjectNN","model":"PointMLP+TAP","rank_in_archive_order":36,"of":77,"metrics":{"Overall Accuracy":"88.5"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2307.14971","atlas_url":"https://app.syntology.ai/?focus=2307.14971","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}