Keep preprocess in the config "True" if running it for the first time. Once you have run it, there is no need to re-process the dataset. The "data/" folder should have the "raw_data" along with the preprocessed and re-arranged dataset i.e. "images/", "questions.lst", "labels.lst", and "templates.lst". If you have these files (updated) then no need to re-do these steps. This can avoided by setting the preprocess param to False in the config file.
conda create -n clevrmath python=3.10 -y
source activate clevrmath
pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt
If running it for the first time, then we need to donwload and preprocess the dataset.
python preprocessing/preprocess_raw_data.py
To get image tensors, make sure to set "get_image_tensors" True in config file.
python run.py