PyTorch implementation of paper "DSNet:Double Strand Robotic Grasp Detection\Network Based on Cross Attention".
The article is currently under submission.
Currently, both the Cornell Grasping Dataset, Jacquard Dataset and Multi Dataset are supported.
- Download the and extract Cornell Grasping Dataset.
-
Download and extract the Jacquard Dataset.
-
We use the setting in here
Training is done by the main.py
script.
Some basic examples:
# Train on Cornell Dataset
python main.py --dataset cornell
# k-fold training
python main_k_fold.py --dataset cornell
Trained models are saved in `output/models` by default, with the validation score appended.
## Visualize
Some basic examples:
```bash
# visualise grasp rectangles
python visualise_grasp_rectangle.py --network your network address
# visualise heatmaps
python visualise_heatmaps.py --network your network address
Code heavily inspired and modified from https://github.com/dougsm/ggcnn.