Code for paper “SSDC-Grasp: A Joint Semantic Segmentation and Depth Completion Approach for Robotic Grasping of Transparent Objects”
Ling Tong, Kechen Song, Member, IEEE, Hongkun Tian, Yi Man, Fiqi Sun, Youwei Chen, Yunhui Yan, and Qinggang Meng, Senior Member, IEEE.
The code of SSNet can be found at code. Or you can use other semantic segmentation methods from MMSegmentation at this.
The video of 4-Dof robotic experiments can be found at this.
The video of 6-Dof robotic experiments can be found at this.
Our test set used in grasp detection experimentson Test set
To train DCNet on the TransCG dataset.
$ python train.py --cfg [Configuration File]
To evaluate DCNet on the TransCG dataset
$ python test.py --cfg [Configuration File]
To visualize the inference results of DCNet on the TransCG dataset. Change the config and checkpoint in inference.yaml, input dirs in sample_inference.py to your path.
$ python sample_inference.py
Using TF-grasp code or other grasp detection methods. Use the result of SSNet and DCNet as input rgb and depth, then predict the grasping poses.
Code heavily inspired and modified from code