Implementation of Rios, A., & Itti, L. (2019, August). Closed-loop memory gan for continual learning. In Proceedings of the 28th International Joint Conference on Artificial Intelligence (pp. 3332-3338). AAAI Press.
Uses pytorch: Torch version 0.4.1
To train CloGAN type:
python3 -i CloGAN_main.py --dataset_name DATASETNAME --dataroot PATH_TO_DATASET --main_out_path PATH_TO_RESULTS --fraction_buff BUFFER_USAGE --save_generated_images
Where the following flags should be modified appropriately:
-
BUFFER_USAGE: Set buffer usage as percentage of dataset (ex: 0.01 = 1% of entire dataset) to store: --fraction_buff 0.01 (ex for svhn 1% or 5%) --fraction_buff 0.01 (ex for emnist 1% or 5%) --fraction_buff 0.01 (ex for mnist 0.1% or 0.5%) --fraction_buff 0.01 (ex for fashion 0.1% or 0.5%)
-
DATASETNAME: Define which dataset to use: --dataset_name mnist --dataset_name fashion --dataset_name emnist --dataset_name svhn
-
PATH_TO_DATASET: Set path to where dataset is stored --dataroot /home/data/ (for example)
-
PATH_TO_RESULTS: Set path to general results folder --main_out_path /home/CloGAN/results/