Fork of google-research/ravens to generate custom dataset.
Step 1. Recommended: install Miniconda with Python 3.7.
curl -O https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh -b -u
echo $'\nexport PATH=~/miniconda3/bin:"${PATH}"\n' >> ~/.profile # Add Conda to PATH.
source ~/.profile
conda initStep 2. Create and activate Conda environment, then install GCC and Python packages.
git clone git@github.com:HHousen/ravens.git
cd ~/ravens
conda create --name ravens python=3.7 -y
conda activate ravens
conda install h5py tqdm
pip install -r requirements.txt
pip install -e .Generate the dataset by running python ravens/demos.py --assets_root=./ravens/environments/assets/ --task=place-red-in-green --mode=train --n=70000. This will create a h5py file called raven_robot_data.h5 with the datasets color (image data), segm (segmentation maps), and num_objects_on_table (number of objects present on the table in the image).
The preview_dataset.py script will cycle though the images in the generated raven_robot_data.h5 dataset file using matplotlib.
- Training Data (first chunk):
python ravens/demos.py --assets_root=./ravens/environments/assets/ --task=place-red-in-green --mode=train --start_seed=-2 --n=11667. - Training Data (second chunk):
python ravens/demos.py --assets_root=./ravens/environments/assets/ --task=place-red-in-green --mode=train --start_seed=23332 --n=11667. - Evaluation/Test Data:
python ravens/demos.py --assets_root=./ravens/environments/assets/ --task=place-red-in-green --mode=test --start_seed=-1 --n=5000. - Merge Training Data Chunk:
python merge.py. - Rename:
mv ravens_robot_data_train_0.h5 ravens_robot_data_train.h5 && mv ravens_robot_data_test_1.h5 ravens_robot_data_test.h5.