Source code for: Class-Incremental Mixture of Gaussians for Deep Continual Learning.
Python version: 3.8
python -m venv venv
. ./venv/bin/activate
python -m pip install --upgrade pip
pip install -r requirements.txt
(optional) tensorboard --logidr runs/run_label
- Original datasets (MNIST, FASHION, SVHN, IMAGENET, CIFAR) will be downloaded automatically.
- Pre-trained features for IMAGENET200 and CIFAR100 can be downloaded from: https://drive.google.com/drive/folders/1gM5rgK-_HS-S1M2LzYRcofXRENieoVbg?usp=sharing
- The extracted features should be put in: pytorch_data/extracted.
- Different configurations for MIX:
python mix_run.py -t params -p all -d all -rl mix-params -dev cuda:0 - Final versions (benchmark/mix/mix_final_exp.py) for MIX:
python mix_run.py -t final -v all -d all -rl mix-final -dev cuda:0 - Baselines:
python mix_run.py -t baselines -b naive er ersb agem der -lr 0.0001 -bn 0 -d all -rl mix-baselines -dev cuda:0python mix_run.py -t baselines -b icarl gss lwf si -lr 0.0001 -bn 1 -d all -rl mix-baselines -dev cuda:0
- Output directory: results.
- [optional] Post-processing (set your paths in the script):
python -m scripts.mix_results_proc - [optional] Tensorboard: http://localhost:6006/