{"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/segmenting-unknown-3d-objects-from-real-depth","title":"Segmenting Unknown 3D Objects from Real Depth Images using Mask R-CNN Trained on Synthetic Data","arxiv_id":"1809.05825","date":"2018-09-16","proceeding":null,"authors":["Michael Danielczuk","Matthew Matl","Saurabh Gupta","Andrew Li","Andrew Lee","Jeffrey Mahler","Ken Goldberg"],"abstract":"The ability to segment unknown objects in depth images has potential to\nenhance robot skills in grasping and object tracking. Recent computer vision\nresearch has demonstrated that Mask R-CNN can be trained to segment specific\ncategories of objects in RGB images when massive hand-labeled datasets are\navailable. As generating these datasets is time consuming, we instead train\nwith synthetic depth images. Many robots now use depth sensors, and recent\nresults suggest training on synthetic depth data can transfer successfully to\nthe real world. We present a method for automated dataset generation and\nrapidly generate a synthetic training dataset of 50,000 depth images and\n320,000 object masks using simulated heaps of 3D CAD models. We train a variant\nof Mask R-CNN with domain randomization on the generated dataset to perform\ncategory-agnostic instance segmentation without any hand-labeled data and we\nevaluate the trained network, which we refer to as Synthetic Depth (SD) Mask\nR-CNN, on a set of real, high-resolution depth images of challenging,\ndensely-cluttered bins containing objects with highly-varied geometry. SD Mask\nR-CNN outperforms point cloud clustering baselines by an absolute 15% in\nAverage Precision and 20% in Average Recall on COCO benchmarks, and achieves\nperformance levels similar to a Mask R-CNN trained on a massive, hand-labeled\nRGB dataset and fine-tuned on real images from the experimental setup. We\ndeploy the model in an instance-specific grasping pipeline to demonstrate its\nusefulness in a robotics application. Code, the synthetic training dataset, and\nsupplementary material are available at https://bit.ly/2letCuE.","url_abs":"http://arxiv.org/abs/1809.05825v2","url_pdf":"http://arxiv.org/pdf/1809.05825v2.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":"segmenting-unknown-3d-objects-from-real-depth","repo_url":"https://github.com/BerkeleyAutomation/sd-maskrcnn","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"segmenting-unknown-3d-objects-from-real-depth","repo_url":"https://github.com/BugBoy13/sd-maskrcnn","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"segmenting-unknown-3d-objects-from-real-depth","repo_url":"https://github.com/2024-MindSpore-1/Code4/tree/main/FasterRCNN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null},{"paper_slug":"segmenting-unknown-3d-objects-from-real-depth","repo_url":"https://github.com/MindSpore-paper-code-2/code400/tree/main/MaskRCNN/maskrcnn_mobilenetv1","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"mindspore","reach":null}],"tasks":[{"task_slug":"clustering","task_name":"Clustering"},{"task_slug":"dataset-generation","task_name":"Dataset Generation"},{"task_slug":"instance-segmentation","task_name":"Instance Segmentation"},{"task_slug":"object-tracking","task_name":"Object Tracking"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"},{"task_slug":"unseen-object-instance-segmentation","task_name":"Unseen Object Instance Segmentation"}],"methods":[{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"mask-r-cnn","method_name":"Mask R-CNN"},{"method_slug":"rpn","method_name":"RPN"},{"method_slug":"roi-align","method_name":"RoIAlign"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[{"slug":"wisdom","name":"WISDOM","full_name":"Warehouse Instance Segmentation Dataset for Object Manipulation"}],"methods_introduced":[],"results":[{"leaderboard":"/sota/unseen-object-instance-segmentation-on-wisdom","task":"Unseen Object Instance Segmentation","dataset":"WISDOM","model":"SD Mask R-CNN","rank_in_archive_order":1,"of":1,"metrics":{"mAP @0.5:0.95":"51.6"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}