license: cc-by-4.0
task_categories:

• object-detection
• robotics
  tags:
• 6d-pose-estimation
• transparency
• megapose
• bop
• synthetic
  pretty_name: Transparent Object Pose Estimation
  size_categories:
• 100K<n<1M

Object Pose Estimation Using Implicit Representation for Transparent Objects

This dataset consists of raw rendered Physically Based Rendering (PBR) data and 3D mesh assets designed for training and fine-tuning pose-estimation models, specifically adapting Megapose for transparent objects. Created by Varun Burde in November 2024 and prepared for distribution in February 2026, the data provides a comprehensive resource for implicit representation research.

Visualization

Below is a sample visualization showing the RGB render, Depth map, and Visibility Mask (stacked sequentially):

Dataset Structure

The dataset is organized into zipped archives for easier accessibility. You will need to unzip these files to access the training data.

Archives

• meshes_and_meta.zip: Contains object meshes (.ply) and global camera parameters.
• train_pbr_xxx_xxx.zip: Split archives containing the rendered training sequences.

Internal File Structure (after unzipping)

The data follows the BOP (Benchmark for 6D Object Pose Estimation) directory structure:

dataset/
├── camera.json                 # Global camera intrinsics/extrinsics
├── meshes/
│   ├── models_info.json        # Metadata about the 3D models
│   ├── obj_000000.ply          # 3D Mesh file for object 0
│   ├── obj_000001.ply          # 3D Mesh file for object 1
│   └── ...
└── train_pbr/
    ├── 000000/                 # Scene ID
    │   ├── scene_camera.json   # Camera parameters for each frame in this scene
    │   ├── scene_gt.json       # Ground truth 6D poses
    │   ├── scene_gt_info.json  # Bounding box and visibility info
    │   ├── rgb/
    │   │   ├── 000000.png      # RGB Image
    │   │   └── ...
    │   ├── depth/
    │   │   ├── 000000.png      # Depth Map
    │   │   └── ...
    │   └── mask_visib/
    │       ├── 000000_000000.png # Visibility mask for object instance 0
    │       └── ...
    ├── 000001/
    └── ...

This work is associated with the research published in Object Pose Estimation Using Implicit Representation for Transparent Objects, available at SpringerView.

In creating this dataset, we utilized the BOP Toolkit for standardized formatting and BlenderProc for the underlying synthetic data generation. This work was supported by the European Union under the project Robotics and advanced industrial production (reg. no. CZ.02.01.01/00/22_008/0004590).

The dataset is distributed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. Users are free to share and adapt the material provided they give appropriate credit to the authors and the associated publication.

If you find this dataset useful, please cite it using:

@InProceedings{10.1007/978-3-031-91569-7_15,
author="Burde, Varun and Moroz, Artem and Zeman, V{\'i}t and Burget, Pavel",
editor="Del Bue, Alessio and Canton, Cristian and Pont-Tuset, Jordi and Tommasi, Tatiana",
title="Object Pose Estimation Using Implicit Representation for Transparent Objects",
booktitle="Computer Vision -- ECCV 2024 Workshops",
year="2025",
publisher="Springer Nature Switzerland",
address="Cham",
pages="226--247",
isbn="978-3-031-91569-7"
}