Download CIFAR-10 and store it in folder ./dataset
.
Generate a number of random patterns corresponding to each data point as npz file by running gen_masks.py
.
Store the npz file in folder ./dataset
.
Run train.py
with default settings.
Finetune the model by changing some training parameters as follows:
args.run_name = 'nf128_na1-4-16_nc32_finetune' \
args.epochs = 2000 \
args.warmup = False \
args.shift_range = 0 \
args.weight_path = 'PATH to initially trained model' \
Please refer to the paper.
Run sample.py
.