This code contains instruction on how to reproduce our results in the paper.

Run the following commands to create a virtual environment, activate it and install the required packages.

conda create -n clipsym python=3.8.18
conda activate clipsym
pip install -r requirements.txt

DENDI dataset can be downloaded from DENDI.

SDRW and LDRS datasets can be downloaded from PMC (password: ldrs2021).

After downloaded, please unzip dataset zip files under the sym_datasets folder. Note, sym_datasets contains a csv file objects.csv, which contains frequent object classes in the DENDI dataset.

python train.py --dataset <DATASET> --method <METHOD> [OTHER PARAMETERS]

The script for training CLIPSym is

python train.py --dataset dendi --method clipsym -eq_up

'--dataset': 'dendi' or 'pmc'
'--method': 'clipsym', 'equisym', 'pmc'
'-eq_up': should be used together with method 'clipsym'. When specified, the upsampler is G-equivariant.

Please check config.py for other hyperparameters. If they are not set, the code will be using the default parameters for reproducing the results reported in the paper.

You may refer https://github.com/ahyunSeo/EquiSym and https://github.com/ahyunSeo/PMCNet for more details about EquiSym and PMCNet.