IC-MI

The IC-MI dataset, introduced by Tejani et al., is part of the Benchmark for 6D Object Pose Estimation (BOP). Let's delve into the details:

  1. Dataset Overview:

    • The IC-MI dataset contains models of two texture-less and four textured household objects.
    • These objects serve as test cases for evaluating 6D object detection and pose estimation methods.
    • The test images showcase multiple object instances with clutter and slight occlusion¹².

  2. Object Categories:

    • Texture-less Objects: Two texture-less objects are included in the dataset.
    • Textured Objects: Four textured household objects are part of the dataset.

  3. Ground-Truth Annotations:

    • The dataset provides training/test RGB-D images that are annotated with the following ground-truth information:
      • 6D object poses: Precise 3D positions and orientations of the objects.
      • 2D bounding boxes: Enclosing the objects in the 2D image plane.
      • 2D binary masks: Indicating the object pixels.

  4. Data Creation:

    • The 3D object models were manually created or reconstructed using systems similar to KinectFusion.
    • Training images were captured by RGB-D/Gray-D sensors or rendered from the 3D models.
    • All test images are real-world captures.

  5. Format and Storage:

    • The datasets are provided in the BOP format.
    • The BOP toolkit expects datasets to be stored in the same folder, with each dataset in a subfolder named after its base name (e.g., "lm," "lmo," "tless").
    • The IC-MI dataset is one of the components of this comprehensive benchmark.

In summary, the IC-MI dataset offers valuable resources for advancing 6D object pose estimation research, particularly in scenarios involving texture-less and textured objects with clutter and mild occlusion¹.

(1) Datasets - BOP: Benchmark for 6D Object Pose Estimation. https://bop.felk.cvut.cz/datasets/. (2) BOP: Benchmark for 6D Object Pose Estimation | SpringerLink. https://link.springer.com/chapter/10.1007/978-3-030-01249-6_2. (3) Sensors | Free Full-Text | Visual Attention and Color Cues for ... - MDPI. https://www.mdpi.com/1424-8220/21/23/8090. (4) undefined. https://bop.felk.cvut.cz/media/data/bop_datasets.