The actor profile images were downloaded from IMDB page. The collection of images can be downloaded here (27GB).
After downloading, un-tar the file to a profile image folder for the next-step usage.

python main.py \
--dataset={dataset} \                  #e.g. "cmdad"
--video_dir={video_dir} \
--anno_path={anno_path} \              #e.g. "resources/annotations/cmdad_anno.csv"
--charbank_path={charbank_path} \      #e.g. "resources/charbanks/cmdad_charbank.json" 
--profile_dir={profile_dir} \
--output_dir={output_dir}

--dataset: choices are cmdad, madeval, and tvad.
--video_dir: directory of video datasets, example file structures can be found in resources/example_file_structures (files are empty, for references only).
--anno_path: path to AD annotations, available in resources/annotations.
--charbank_path: path to external character banks, available in resources/charbanks.
--profile_dir: directory of actor profile images.
--output_dir: directory to save output csv.

InsightFace: https://github.com/deepinsight/insightface