- Make sure you have a working version of nvidia-docker or docker installed on your computer.
- Clone this gqcnn repository to a directory of your choice:
$WORKING_DIR/gqcnn
- Create a directory to store your results in, e.g.
$RESULTS
. - Download the trained model
from the SharePoint and extract the zip file in
$WORKING_DIR/gqcnn/.
. - Run
./build.sh
from$WORKING_DIR/gqcnn/
. Choose 1 if you have nvidia-docker and would like to use the GPU, 0 otherwise. - Set the variables
DATA_PATH
andEXPER_PATH
ingqcnn/run_docker.sh
to your$DATA_DIR
and$RESULTS
directories.$DATA_DIR
should be pointing to the directory where your DexNet datasets are stored. - Run
./run_docker.sh
from$WORKING_DIR/gqcnn/
.
In order to evaluate a dataset, run
python3 tools/detailed_analysis.py GQCNN-2.0_benchmark $DATASET_DIR --output_dir $DATASET_OUTPUT_DIR
For example, if you want to evaluate a recreated PerfectPredictions dataset, run:
python3 tools/detailed_analysis.py GQCNN-2.0_benchmark Recreated_grasps/tensors/ --output_dir Recreated_grasps
The logfile and the images with prediction values are stored in $RESULTS/$DATASET_OUTPUT_DIR/
.
The gqcnn Python package is for training and analysis of Grasp Quality Convolutional Neural Networks (GQ-CNNs). It is part of the ongoing Dexterity-Network (Dex-Net) project created and maintained by the AUTOLAB at UC Berkeley.
Please see the docs for installation and usage instructions.
If you use any part of this code in a publication, please cite the appropriate Dex-Net publication.