10th International Workshop on Recovering 6D Object Pose (R6D)

Program

13:00	Opening: Martin Sundermeyer (Google)
13:10	Talk 1: Hao-Shu Fang (MIT)
13:40	Talk 2: Maximilian Durner (DLR)
14:10	Coffee break at posters (workshop and invited conference papers)
15:00	Talk 3: Agastya Kalra & Vahe Taamazyan (Intrinsic)
15:30	Talk 4: Sergey Levine (Physical Intelligence)
16:00	Panel Discussion
16:20	BOP Challenge 2025: Martin Sundermeyer, Tomas Hodan, Lukas Ranftl, Junwen Huang (all authors)
17:00	End of workshop

BOP Challenge 2025

BOP 2017–2024 summary: To measure the progress in the field of object pose estimation, we created the BOP benchmark in 2017 and have been organizing challenges on the benchmark datasets together with the R6D workshops since then. We have witnessed major improvements in model-based object pose estimation: the accuracy has increased by more than 50% since 2019, competitive results have been achieved when training only on synthetic images, and the top 2024 method for unseen objects (CAD models not available at training) achieves the same level of accuracy as the best 2022 method for seen objects (BOP'24 report). In BOP 2024, we also introduced the new BOP-H3 datasets on which we additionally evaluate model-free methods, i.e., instead of a CAD model, only reference images of the target object are available for object onboarding. While the model-free problem is most relevant for open-world applications, industrial robotics applications are usually model-based and face different challenges such as strong occlusions, reflections, clutter and symmetries while requiring high precision and speed.

BOP 2025: Together with the workshop, we organize the BOP Challenge 2025 on model-based and model-free 2D/6D object detection on BOP-Classic, BOP-H3 and new BOP-Industrial datasets (XYZ-IBD, ITODD-MV, IPD). The new BOP-Industrial datasets focus on industrial objects in bin-picking and other industry-relevant scenarios (sample images are below). BOP-Industrial datasets enable evaluating both single-view and multi-view methods and include a variety of sensors and modalities to determine the best setup for robotics applications. The development of high-accuracy pose estimation methods is crucial for the progression of industrial automation and present challenges to current methods.

Join the challenge

Associated BPC challenge: BOP Challenge 2025 is associated with the OpenCV Perception Challenge for Bin-picking (BPC), which offers $60,000 in prizes and is co-organized by OpenCV, Intrinsic, BOP, Orbbec, and University of Hawaii at Manoa. Phase 1 of the OpenCV challenge is evaluated on the IPD dataset which is also included in BOP Challenge 2025 – participants can evaluate the pose accuracy in BOP before submitting a full docker image to BPC.

Dates

Early-bird submission deadline for the BOP Challenge 2025: June 1, 2025 (11:59PM UTC)
Final Submission deadline for the BOP Challenge 2025: October 1, 2025 (11:59PM UTC)

In-proceedings paper submission deadline: June 30, 2025 (11:59PM PST)
In-proceedings paper acceptance notification: Jul 14, 2025
In-proceedings paper camera-ready version: Aug 14, 2025

Non-proceedings paper submission deadline: October 1, 2025 (11:59PM PST)
Non-proceedings paper acceptance notification: October 5, 2025

Workshop date: October 19, 2025