IC-BIN

The IC-BIN dataset was introduced by Doumanoglou et al. as part of their research on recovering 6D object pose and predicting next-best-view in the crowd¹². This dataset is specifically designed to address the challenges posed by reflective objects in robotic bin-picking scenarios.

Here are the key details about the IC-BIN dataset:

  1. Purpose: The IC-BIN dataset aims to facilitate research in 6D object pose estimation and active vision techniques for reflective objects commonly encountered in bin-picking applications.

  2. Contents:

    • The dataset comprises multiple objects stacked in a bin.
    • It includes three scenes, each containing two objects from the IC-MI dataset.
    • These scenes were recorded from different viewpoints to evaluate object pose estimation methods.

  3. Challenges:

    • Reflective objects are often texture-less and cannot be reliably recognized using classic techniques based on local descriptors.
    • The high glossiness of these objects can introduce fake edges in RGB images and lead to inaccurate depth measurements, especially in cluttered bin scenarios.

  4. Data Annotation:

    • For each scene, the dataset provides monochrome/RGB images and depth maps captured from sampled view spheres around the scene.
    • These images and maps are annotated with accurate 6D poses of visible objects and an associated visibility score.
    • Ground truth depth maps were captured using a high-cost Ensenso camera with objects coated in anti-reflective scanning spray.

  5. Utility and Evaluation:

    • Researchers can use the IC-BIN dataset to evaluate the performance of depth fusion algorithms.
    • Evaluation results highlight the difficulty of handling highly reflective objects, especially in challenging cases with degraded depth data quality, severe occlusions, and cluttered scenes.

(1) ROBI: A Multi-View Dataset for Reflective Objects in Robotic Bin-Picking. https://arxiv.org/pdf/2105.04112v1. (2) Datasets - BOP: Benchmark for 6D Object Pose Estimation. https://bop.felk.cvut.cz/datasets/. (3) ROBI: A Multi-View Dataset for Reflective Objects in Robotic Bin-Picking. https://ar5iv.labs.arxiv.org/html/2105.04112. (4) undefined. https://www.trailab.utias.utoronto.ca/robi.