 In this folder is contained the source code to run all the experiment of the paper "Generative Negative Replay for Continual Learning" submitted to ICLR 2022. All the content of this folder is highly confidential and cannot be shared to people other than the reviewer of the paper. 
 
 The folder is structured in 3 sub-folders, each of them containing the code of a particular scenario: 
 - CORe50NC -> Code for the experiments on the CORe50 NC scenario (+ the code for the generative replay for CORe50 NIC)
 - CORe50NIC -> Code for the experiments on the CORe50 NIC scenario (except the code for the generative replay for CORe50 NIC)
 - ImageNet1000 -> Code for the experiments on the ImageNet-1000 scenario
 
 First download the pretrained models and the class ordering from here: https://drive.google.com/file/d/1tZqIIUIrWDnbsLZjXtQFssS_uyxFam9d/view?usp=sharing

 Unzip the file and copy the content of the subfolders in the downloaded file in the respective subfolders in this folder.
 
 To run the code you may first want to create the conda environments using the ".yml" files provided inside the respective folders. Note that we suppose a Linux server with at least 1 GPU. Modify the files (especially the cudatoolkit version) to suit your system's hardware. 
 
 To run an experiment just run "python <name-of-the-experiment>.py 
 
 By default it will run the first run with the first seed. To run the experiments with another seed follow the instruction contained in each folder. 
 
 You may want to download the datasets first. Follow the link on the paper and download them before running the code. You may want to change the path where the code search for your data. Just search in the source python file "you/path/here" and substitute with your path. 
