The code will automatically download most datasets, as per file src/utils/dataset_factory.py. You can check the file to see which address are being used to download the datasets.
The only dataset that can't be automatically downloaded and configured by our scripts is the Office31 Dataset.
Please, manually download the dataset from https://www.hemanthdv.org/officeHomeDataset.html and unzip in the folder ./data/office_home. (Create the folder, if it doesn't exist. If you run our scripts, it will automatically create this folder for you).
Install the required libraries in the requirements.txt file. You can do so by using the following command:
pip install -r requirements.txt
You can also install it via a conda environment using the file conda-environment.yaml:
conda env create -f environment.yml
You might get an ImportError with the dataset Kinetics400 from torchvision. If this happens in your environment, you will need to edit the corresponding import line on othe torchvision source-code itself, as a work-around. The path in a miniconda linux environment is /home/USER/miniconda3/envs/CDCL/lib/python3.9/site-packages/avalanche/benchmarks/datasets/torchvision_wrapper.py.
You can re-run all experiments by executing the file cdcl_paper.sh. You might need to edit its hidden characters (space character) if you are running it in Windows or Linux.