The basis of many object manipulation algorithms is RGB-D input. Yet, commodity RGB-D sensors can only provide distorted depth maps for a wide range of transparent objects due light refraction and absorption. To tackle the perception challenges posed by transparent objects, we propose TranspareNet, a joint point cloud and depth completion method, with the ability to complete the depth of transparent objects in cluttered and complex scenes, even with partially filled fluid contents within the vessels. To address the shortcomings of existing transparent object data collection schemes in literature, we also propose an automated dataset creation workflow that consists of robot-controlled image collection and vision-based automatic annotation. Through this automated workflow, we created Transparent Object Depth Dataset (TODD), which consists of nearly 15000 RGB-D images. Our experimental evaluation demonstrates that TranspareNet outperforms existing state-of-the-art depth completion methods on multiple datasets, including ClearGrasp, and that it also handles cluttered scenes when trained on TODD.
Dataset: https://dataverse.scholarsportal.info/dataset.xhtml?persistentId=doi:10.5683/SP3/ZJJAJ3
This repository provides:
- Point cloud completion module
- Depth completion module
- Dataset collection utility tool
git clone https://github.com/pairlab/TranspareNet.git
pip install -r requirements.txt
TranspareNet inference:
python inference_transparenet.py
Point cloud completion training
python pccRunner.py
Convert dataset depth images to objects point clouds
python grnet_point_cloud_completion/datasets/img2pcd.py
Convert predicted point clouds to sparse depth estimation
pyhton grnet_point_cloud_completion/datasets/pcd2img.py
Depth completion training
python tools/train_franka.py
Requirement:
- frankapy
- Franka Emika panda robot
- Intel Realsense D435i camera
- Intel Realsense ROS
Our point cloud completion network is based on GRNet, and our depth completion network is based on DMLRN