SenseShift6D: Multimodal RGB-D Benchmarking for Robust 6D Pose Estimation across Environment and Sensor Variations

📷 Dataset Information

SenseShift6D is an RGB-D dataset for evaluating object pose estimation under variations in illumination and camera sensor settings. It contains data for six household objects captured across 5 illumination levels, 13 RGB exposures, 9 RGB gains, auto-exposure and 4 depth-capture modes, with accurate 6D pose annotations (≈ 2.32 mm).

🎛️ Configuration

Each split was captured under the following configurations:

Split	Brightness (%)	RGB exposure (µs)	RGB gain	Depth-capture mode
Train	5, 25, 50, 75, 100	Auto, 2, 4, 19, 78, 312, 1250, 5000, 10000	Auto, 0, 32, 64, 96, 128	Default, High Accuracy, High Density, Medium Density
Test	5, 25, 50, 75, 100	Auto, 9, 39, 156, 625, 2500	Auto, 16, 48, 80, 112	Default, High Accuracy, High Density, Medium Density

🗂️ Structure

The dataset follows the BOP format, with sensor-specific subfolders.

SenseShift6D/
└── dataset/
    ├── train/
    │   └── B{x}/                       # Brightness condition, e.g. B5 or B75
    │       └── {scene_id}/             
    │           ├── rgb/
    │           │   └── E{x}G{x}/       # RGB exposure/gain setting, e.g. E19G32
    │           │       └── *.png
    │           ├── depth/
    │           │   └── {preset_num}/   # Depth preset: 0, 1, 2, or 3
    │           │       └── *.png
    │           ├── mask/
    │           │   └── *.png           # Full object masks
    │           ├── mask_visib/
    │           │   └── *.png           # Visible object masks
    │           ├── scene_camera.json   # Camera parameters
    │           ├── scene_gt.json       # Ground-truth poses
    │           └── scene_gt_info.json  # Bounding boxes and visibility information
    ├── test/
    │   └── B{x}/
    │       └── ...                     # Same structure as train
    ├── models/                         # 3D CAD models and information
    │   ├── obj_000001.ply
    │   ├── obj_000002.ply
    │   ├── ...                         
    │   └── models_info.json            
    └── models_eval/
        └── ...                         # Same structure as models

The depth preset nums correspond to the following capture modes:

Folder	Depth capture mode
0	Default
1	High Accuracy
2	High Density
3	Medium Density

📚 Citation

If you use SenseShift6D in your work, please cite the following paper:

@inproceedings{Han_2027_WACV,
title={SenseShift6D: Multimodal RGB-D Benchmarking for Robust 6D Pose Estimation across Environment and Sensor Variations},
author={Han, Yegyu and Yoon, Taegyoon and Woo, Dayeon and Kim, Sojeong and Kim, Hyung-Sin},
booktitle={Proceedings of the Winter Conference on Applications of Computer Vision (WACV)},
year={2027},
}