In this document, we provide instructions on how to prepare data for training/testing our prediction model on SimB Dataset.
You could download the SimB dataset using the following script:
gdown --id 1ucTEMTyzWS1uaknrFU3gTpE3KnNlKAIf -O data/simb.zip
unzip data/simb.zip -d data/
If you are unable to run the script above, you can also try to download using the link, and unzip it to have the following file structures:
data/simb/train.hkl
data/simb/test.hkl
After you have download the hickle files, you need to use the following command to process the data:
python tools/prepare_billiard.py --split train
python tools/prepare_billiard.py --split test
Then, you can evaluate our pre-trained models and train your own models. To evaluate the pretrained models, download them using the following script or the following links for RPCIN.
gdown --id 1vbJWlLCdT6GqTqry61TB3eEOGtDg9q-J -O outputs/phys/simb/rpcin.zip
unzip outputs/phys/simb/rpcin.zip -d outputs/phys/simb/
Then you can run evaluation using:
sh scripts/test_pred.sh simb rpcin rpcin ${GPU_ID}
To train the model, you can simply run the following command:
# For RPCIN
python train.py --cfg configs/simb/rpcin.yaml --gpus ${GPU_ID} --output ${OUTPUT_NAME}
In case you want to generate your customized version of the simulated billiard dataset, read the following file tools/gen_billiard.py. Our SimB dataset is generated using the following command (but note that the dataset may be slightly different because of randomness):
python tools/gen_billiard.py # script used for generating simb