Minimal implementation of DDPM w/ CFG from scratch.
Trained on English typefaces dataset from https://www.kaggle.com/datasets/killen/bw-font-typefaces
Capable of conditional generation of varied English Typefaces.
Put each class image in a seperate directory, and update path of parent directory in training script:
args.dataset_path = <path to dataset>
args.num_classes = <number of classes in dataset>
args.run_name = <some cool name>
And everything should hopefully work fine.
python generations.py --run_name DDPM_cond1 --plot_save_path /tmp --string demonstration -cfg_scale 5
