Datasets:
license: cc-by-nc-sa-4.0
tags:
- 6d-pose-estimation
size_categories:
- 10K<n<100K
dataset_info:
features:
- name: scene_id
dtype: string
- name: image_id
dtype: string
- name: camera_type
dtype: string
- name: image
dtype: image
- name: depth
dtype: image
- name: width
dtype: int64
- name: height
dtype: int64
- name: split
dtype: string
- name: labels
dtype: string
- name: camera_params
dtype: string
- name: source_image_id
dtype: string
configs:
- config_name: default
data_files:
- split: test
path: data/*
pretty_name: CHIP
CHIP: A multi-sensor dataset for 6D pose estimation of chairs in industrial settings
Introduction
Accurate 6D pose estimation of complex objects in 3D environments is essential for effective robotic manipulation. Yet, existing benchmarks fall short in evaluating 6D pose estimation methods under realistic industrial conditions, as most datasets focus on household objects in domestic settings, while the few available industrial datasets are limited to artificial setups with objects placed on tables. To bridge this gap, we introduce CHIP, the first dataset designed for 6D pose estimation of chairs manipulated by a robotic arm in a real-world industrial environment. CHIP includes seven distinct chairs captured using three different RGBD sensing technologies and presents unique challenges, such as distractor objects with fine-grained differences and severe occlusions caused by the robotic arm and human operators. CHIP comprises 77,811 RGBD images annotated with ground-truth 6D poses automatically derived from the robot's kinematics, averaging 11,115 annotations per chair. We benchmark CHIP using three zero-shot 6D pose estimation methods, assessing performance across different sensor types, localization priors, and occlusion levels. Results show substantial room for improvement, highlighting the unique challenges posed by the dataset.
Dataset Summary
- Number of images: 77,811 RGBD images
- Number of object classes: 7 distinct chair models
- Sensors used: Intel RealSense D435, Intel RealSense L515, Stereo Labs ZED
- Annotations: Ground-truth 6D poses derived from robot kinematics (~11,115 annotations per chair)
- Occlusion levels: No occlusions, moderate occlusions
Object Classes
CHIP includes seven high-quality chair models from Andreu World, covering a variety of structures and materials.
Frameonly designs:
- 000001: Smile si0325 — Andreu World link
- 000003: Carlotta si0991 — Andreu World link
- 000006: Carola so0903 — Andreu World link
- 000007: Rizo so2043 — Andreu World link
Solid-wood designs:
- 000002: Happy si0374 — Andreu World link
- 000004: Duos si2750 — Andreu World link
- 000005: Rdl si7291 — Andreu World link
Data Fields
- scene_id: Unique identifier for each scene in the dataset (BOP format).
- image_id: Unique identifier for each image within a scene and camera type (BOP format).
- camera_type: Type of camera used to capture the image (e.g., 'zed', 'rs_l515', 'rs_d435').
- image: RGB image captured by the specified camera.
- depth: Depth image corresponding to the RGB image, captured by the specified camera.
- width: Width of the image in pixels.
- height: Height of the image in pixels.
- split: Dataset split to which the image belongs (e.g., 'test_no_occlusions', 'test_moderate_occlusions').
- source_image_id: Original image identifier from the CHIP dataset.
- labels: JSON string containing object annotations, including 6D poses and visibility information.
- camera_params: JSON string containing intrinsic and extrinsic camera parameters for the specified camera.
Example of Usage
To get started quickly, you can run the example script included in this repository:
This script shows how to load CHIP samples from the dataset, automatically download the required models, and reconstruct a 3D point cloud from the RGB-D image. It then overlays the lifted point cloud with the corresponding object mesh using the ground-truth 6D pose, allowing you to visually inspect the alignment in 3D.
This minimal example demonstrates how to:
- Load RGB-D data and camera parameters
- Lift depth into a point cloud
- Load the object mesh for a given chair
- Apply the ground-truth 6D pose
- Visualize everything together in 3D
Run the script with:
python example_usage.py
and a 3D viewer will open, showing the reconstructed scene and the posed chair model.
Uses
The CHIP dataset can be used for a wide range of research tasks, including:
- Benchmarking 6D pose estimation in realistic industrial environments
- Evaluating robustness to challenging conditions, such as occlusions, clutter, and visually similar distractor objects
- Studying cross-sensor generalization across LiDAR-based, passive-stereo, and active-stereo RGB-D sensors
- 6D pose tracking during robotic manipulation, including motion sequences with dynamic occlusions
Citation
If you find CHIP useful for your work please cite:
@inproceedings{nardon2025chip,
title={CHIP: A multi-sensor dataset for 6D pose estimation of chairs in industrial settings},
author={Nardon, Mattia and Mujika Agirre, Mikel and González Tomé, Ander and Sedano Algarabel, Daniel and Rueda Collell, Josep and Caro, Ana Paola and Caraffa, Andrea and Poiesi, Fabio and Chippendale, Paul Ian and Boscaini, Davide},
booktitle={British Machine Vision Conference (BMVC)},
year={2025}}
Acknowledgement
This work was supported by the European Union's Horizon Europe research and innovation programme under grant agreement No. 101058589 (AI-PRISM).
Partners
Contact
For any questions regarding the dataset, please contact Mattia Nardon (mattia.nardon@fbk.eu).
Homepage link: https://tev-fbk.github.io/CHIP/
Paper link: https://arxiv.org/abs/2506.09699