BOP (Benchmark for 6D Object Pose Estimation)

A public benchmark and yearly challenge for 6D object pose estimation, maintained by the Czech Technical University.

BOP was introduced by Tomáš Hodaň and colleagues at ECCV 2018 and is maintained by the Czech Technical University in Prague. It unifies multiple object pose datasets — LM-O, T-LESS, YCB-V, and others — into a common format, provides a shared error function that handles pose ambiguity for symmetric objects, and runs an online evaluation system; it has held challenges in 2019, 2020, and 2022 through 2025, paired with workshops like R6D. The task has expanded from “seen objects” to “unseen objects”: the 2023 edition required methods to handle new objects quickly from just a CAD model, 2024 added a model-free task using only a reference video plus the BOP-H3 dataset, and 2025 added the BOP-Industrial dataset for industrial scenes. The 2024 report notes that 2D detection of unseen objects still lags detection of seen objects by about 35%, the main bottleneck. Methods including FoundationPose, MegaPose, and SAM-6D are all compared here.

ExampleA newly proposed zero-shot pose estimation method is run on the BOP-Classic-Core dataset, and its results are uploaded to the BOP online evaluation system for an AR score, which is then compared against methods like FoundationPose on the leaderboard.

Also called

BOP Challenge

Related

6D Object Pose Estimation · FoundationPose · MegaPose · SAM-6D · YCB Object and Model Set · Average Distance of Model Points

Sources

BOP: Benchmark for 6D Object Pose Estimation（官网） (Chinese)
BOP: Benchmark for 6D Object Pose Estimation (arXiv 1808.08319, ECCV 2018)
BOP Challenge 2024 on Model-Based and Model-Free 6D Object Pose Estimation (arXiv 2504.02812)

As of

2025-11