Awesome

YCB-Sight: A visuo-tactile dataset for object understanding

cover

YCB-Sight is a visuo-tactile dataset including the simulated and real data from a GelSight tactile sensor and Kinect Azure RGB-D camera on the YCB dataset.

Dataset

You can find the whole dataset here, or download partial data below

YCBSight-Sim

cover

Simulated tactile and depth data with Taxim and pyrender

Object Name	Size (MB)	Link
002_master_chef_can	64.3	[Link]
003_cracker_box	63.2	[Link]
004_sugar_box	61.2	[Link]
005_tomato_soup_can	63.8	[Link]
006_mustard_bottle	63.8	[Link]
007_tuna_fish_can	63.2	[Link]
008_pudding_box	61.5	[Link]
009_gelatin_box	60.3	[Link]
010_potted_meat_can	62.8	[Link]
011_banana	63.7	[Link]
012_strawberry	64.2	[Link]
013_apple	63.4	[Link]
014_lemon	63.4	[Link]
017_orange	63.2	[Link]
019_pitcher_base	64.5	[Link]
021_bleach_cleanser	62.6	[Link]
024_bowl	65.1	[Link]
025_mug	64.2	[Link]
029_plate	66.3	[Link]
035_power_drill	64.7	[Link]
036_wood_block	60.1	[Link]
037_scissors	64.2	[Link]
042_adjustable_wrench	64.7	[Link]
043_phillips_screwdriver	63.9	[Link]
048_hammer	64.1	[Link]
055_baseball	63.5	[Link]
056_tennis_ball	63.4	[Link]
072-a_toy_airplane	65.5	[Link]
072-b_toy_airplane	63.8	[Link]
077_rubiks_cube	61.3	[Link]

YCBSight-Real

cover

Collected tactile and depth data from real world experiments

Object Name	Size (GB)	Link
002_master_chef_can	0.97	[Link]
004_sugar_box	1.15	[Link]
005_tomato_soup_can	1.09	[Link]
010_potted_meat_can	1.09	[Link]
021_bleach_cleanser	1.23	[Link]
036_wood_block	1.02	[Link]

Data directory format

YCBSight-Sim
├── obj1
│   ├── gt_contact_mask
│   │   ├── <idx>.npy
│   │   └── ...
│   ├── gt_height_map
│   │   ├── <idx>.npy
│   │   └── ...
│   ├── gelsight
│   │   ├── <idx>.jpg
│   │   └── ...
│   ├── pose.txt
│   ├── depthCam.npy
│   └── depthCam.pdf
├── obj2
└── ...

YCBSight-Real
├── obj1
│   ├── gelsight
│   │   ├── gelsight_<idx>_<timestamp>.jpg
│   │   └── ...
│   ├── depth
│   │   └── depth_0_<timestamp>.tif
│   ├── pc
│   │   └── pc_0_<timestamp>.npy
│   ├── rgb
│   │   ├── rgb_<idx>_<timestamp>.jpg
│   │   └── ...
│   ├── robot.csv
│   ├── tf.json
│   └── obj1.mp4
├── obj2
└── ...

Dependencies

The visualization and data processing are implemented in python3 and require numpy, scipy, matplotlib, cv2.

To install dependencies: pip install -r requirements.txt.

Data Visualization

scripts/lookup_mapping/lookup.py reconstructs the height maps from the tactile readings. Here are several parameters to set:
- path2model: the path to the directory storing the YCBSight-Real and/or YCBSight-Sim
- sim: True/False, visualize whether the simulated data or real data
- obj: specify a certain object's data to visualize, or set to None to visualize all the data

Local Shape Reconstruction from Touch with Lookup Table

scripts/data_visualization/data_visualizer.py visualize the data in YCB-Sight dataset.

Local Shape Reconstruction from Touch with FCRN network

Please refer to this repo (pytorch version) and this repo (tensorflow version).

License

This dataset is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License, with the accompanying processing code licensed under the MIT License.

Citation

If you use YCB-Sight dataset in your research, please cite:

@article{suresh2021efficient,
  title={Efficient shape mapping through dense touch and vision},
  author={Suresh, Sudharshan and Si, Zilin and Mangelson, Joshua G and Yuan, Wenzhen and Kaess, Michael},
  journal={arXiv preprint arXiv:2109.09884},
  year={2021}
}