1st ICCV Workshop and Challenge on Category-Level Object Pose Estimation in the Wild

Sep. 15th, 2025

Summary

This workshop addresses the critical problem of category-level object pose estimation and its applications within complex robotic manipulation scenarios. Pose estimation, a fundamental challenge in both 3D computer vision and robotics perception, involves accurately determining an object's complete 6-degree-of-freedom (6DoF) pose, comprising its 3D rotation and translation. Our workshop specifically focuses on advancing category-level pose estimation methods under realistic and demanding robotic manipulation settings, particularly emphasizing articulated objects, dynamic environments with potential human-object interactions, and objects subject to severe occlusions and partial visibility.

Benchmarks

To facilitate rigorous evaluation and stimulate innovation, the workshop will include two benchmarks as the following.

Category-Level Pose Estimation in Complex Real-World Scenarios

This track aims to benchmark current methods of category-level pose estimation using diverse, challenging datasets representative of real-world robotics scenarios. Participants will be encouraged to develop robust, generalizable, and computationally efficient algorithms. We will utilize existing large-scale benchmark datasets, such as PACE and Omni6DPose, together with our another curated dataset specifically collected for real-world robotic manipulation scenarios with UR-5, including articulated objects, dynamic object-camera interactions, within-hand manipulation scenarios, and complex occlusions.

Pose Estimation Evaluation Through Downstream Robotic Manipulation Tasks

The second track assesses the practical efficacy of pose estimation methods by integrating them directly into downstream robotic manipulation tasks. The 6DoF pose has been established as a reliable and invariant representation, significantly bridging the simulation-to-real-world transfer gap. In this challenge, we will employ the ManiSkill simulation environment, featuring tasks such as opening microwaves, drawer manipulation, pick-and-place operations, in-hand rotation, and water pouring. Participants will be provided with an oracle manipulation policy pre-trained using ground-truth 6DoF poses, achieving near-perfect performance. Competing methods will supply pose estimation inputs to this policy, and their performance will be directly measured through manipulation task success rates. Thus, higher accuracy in pose estimation is directly correlated with increased task success. We will also compare their inference speed as real-time performance is often a requirement for robotic manipulation.

Workshop Objectives

Foster a deeper understanding of the challenges associated with category-level object pose estimation in varied environments.
Encourage the development and sharing of innovative techniques that ensure robust model performance on large-scale datasets.
Promote community-wide collaboration to continually refine benchmarks and methodologies, advancing the field of computer vision.

Expected Benefits

Gaining insights into the latest research and technological advancements in category-level pose estimation.
Engaging with leading experts to discuss the current challenges and future directions of the field.
Accessing new tools and frameworks that facilitate the development of more effective and adaptable pose estimation models.

By bridging cutting-edge advancements in category-level pose estimation with realistic robotic manipulation tasks, this workshop aims to foster interdisciplinary collaboration, inspire novel methodologies, and ultimately advance the capabilities of robotic agents operating autonomously in dynamic, unstructured, and complex real-world environments.

Call for papers

We welcome submissions of full papers (up to 8 pages) and short papers (up to 4 pages), excluding references and appendices. All submissions must adhere to the ICCV 2025 format and will undergo peer review by domain experts. Accepted papers will be featured either as oral presentations or posters during the workshop. Topics of interest include, but are not limited to:

Category-level 6D object pose estimation and tracking
Open-vocabulary object segmentation and 6D pose estimation
6D pose estimation for optically challenging objects, such as transparent or reflective surfaces
Pose-aware robotics: leveraging object pose in autonomous robotic systems

We encourage novel research that pushes the boundaries of category-level object pose estimation, with a focus on real-world challenges and applications.