| CARVIEW |
TransCG:
TransCG is the first large-scale real-world dataset for transparent object depth completion and grasping.
We provide the original RGB-D images as well as the refined ground-truth depth images in the dataset.
The transparent object pose, transparent object mask, the ground-truth surface normals and transparent object models are also provided.
DFNet: DFNet is a robust, efficient and effective network for depth completion. The detailed source code and the pretrained models of our our depth completion network DFNet is released at this repository.
Download
transcg-data-1.zip (scene 1-10) [13.9 GB] [Google] [Baidu (code: 3n0d)]
transcg-data-2.zip (scene 11-20) [13.0 GB] [Google] [Baidu (code: umim)]
transcg-data-3.zip (scene 21-30) [15.1 GB] [Google] [Baidu (code: g63o)]
transcg-data-4.zip (scene 31-40) [15.0 GB] [Google] [Baidu (code: 6tk3)]
transcg-data-5.zip (scene 41-50) [13.7 GB] [Google] [Baidu (code: kp14)]
transcg-data-6.zip (scene 51-60) [14.8 GB] [Google] [Baidu (code: jmsu)]
transcg-data-7.zip (scene 61-70) [14.9 GB] [Google] [Baidu (code: fqv9)]
transcg-data-8.zip (scene 71-80) [15.4 GB] [Google] [Baidu (code: equb)]
transcg-data-9.zip (scene 81-90) [15.9 GB] [Google] [Baidu (code: msc7)]
transcg-data-10.zip (scene 91-100) [15.5 GB] [Google] [Baidu (code: ncg0)]
transcg-data-11.zip (scene 101-110) [14.2 GB] [Google] [Baidu (code: wntp)]
transcg-data-12.zip (scene 111-120) [15.0 GB] [Google] [Baidu (code: in9v)]
transcg-data-13.zip (scene 121-130) [15.4 GB] [Google] [Baidu (code: rmed)]
Real robot demo
Check out our source codes!
Many objects
Robust to large variance in scenes
Various scenes
Closer to real-world environments
Format
1. Place all the data in the following structure, and there are 130 scenes in total.
|-- transcg
|-- scene1/
|-- scene2/
|-- ... ...
|-- scene130/
|
|-- models
| |-- 0-bottle1.ply # Model of object 0 (bottle1)
| |-- 1-bottle2.ply # Model of object 1 (bottle2)
| |-- ... ...
| `-- 59-wash3.ply # Model of object 59 (wash3)
|
|-- camera_intrinsics # Camera intrinsics
| |-- 1-camIntrinsics-D435.npy # Camera intrinsics for D435 camera
| `-- 2-camIntrinsics-L515.npy # Camera intrinsics for L515 camera
|
|-- T_camera2_camera1.npy # The pose of L515 camera w.r.t. D435 camera
|
`-- metadata.json # Meta-data for the dataset
2. Detail structure of each scene (take scene1 as an example)
|-- scene1
|-- 0 # Perspective 0 (if exists)
| |-- corrected_pose # The refined pose of the transparent objects (w.r.t camera)
| | |-- 0.npy # The pose of object 0 w.r.t. D435 camera
| | |-- 32.npy # The pose of object 32 w.r.t. D435 camera
| | |-- 42.npy # The pose of object 42 w.r.t. D435 camera
| | `-- 47.npy # The pose of object 47 w.r.t. D435 camera
| |
| |-- rgb1.png # RGB image of perspective 0 (D435 camera) (if exists)
| |-- depth1.png # Raw depth image of perspective 0 (D435 camera) (if exists)
| |-- depth1-gt.png # Refined ground-truth depth image of perspective 0 (D435 camera) (if exists)
| |-- depth1-gt-mask.png # Transparent object mask for depth image of perspective 0 (D435 camera) (if exists)
| |-- depth1-gt-sn.png # Ground-truth surface normals of perspective 0 (D435 camera) (if exists)
| |-- rgb2.png # RGB image of perspective 0 (L515 camera) (if exists)
| |-- depth2.png # Raw depth image of perspective 0 (L515 camera) (if exists)
| |-- depth2-gt.png # Refined ground-truth depth image of perspective 0 (L515 camera) (if exists)
| |-- depth2-gt-mask.png # Transparent object mask for depth image of perspective 0 (L515 camera) (if exists)
| `-- depth2-gt-sn.png # Ground-truth surface normals of perspective 0 (L515 camera) (if exists)
|
|-- 1 # Perspective 1 (if exists)
| |-- same structure as 0
|
|-- ... ...
|
|-- 239 # Perspective 239 (if exists)
| |-- same structure as 0
|
`-- metadata.json # Meta-data for the scene.
License
All data, labels, code and models belong to the graspnet team, MVIG, SJTU and are licensed under a Creative Commons Attribution 4.0 Non Commercial License (BY-NC-SA). They are freely available for free non-commercial use, and may be redistributed under these conditions. For commercial queries, please drop an email at fhaoshu at gmail.com and cc lucewu at sjtu.edu.cn .
Check it out!
Explore our website for more details.
Please cite our paper if it helps your research:
@ARTICLE{fang2022transcg,
author={Fang, Hongjie and Fang, Hao-Shu and Xu, Sheng and Lu, Cewu},
journal={IEEE Robotics and Automation Letters},
title={TransCG: A Large-Scale Real-World Dataset for Transparent Object Depth Completion and A Grasping Baseline},
year={2022},
volume={},
number={},
pages={1-8},
doi={10.1109/LRA.2022.3183256}}
Copyright © 2021 Machine Vision and Intelligence Group, Shanghai Jiao Tong University.