AE tutorial

Trained AE model by cifar10.

Compared to the original implementation, we have added batchnorm and changed the compression dimension.

paper

https://arxiv.org/abs/2201.03898

FAA

Feed-forward architecture

MNIST

epoch: 189
test_loss          0.010081866756081581

CIFAR10

epoch: 48
test_loss           0.03823928162455559

CA

Convolutional auto encoder architecture

MNIST

epoch: 68
test_loss          4.327146598370746e-05

CIFAR10

epoch:
epoch: 86
test_loss          0.001922393450513482

Usage

If you test this script, please remove below in Dockerfile.

WORKDIR /root
COPY setup_vim/ ./setup_vim/
RUN ./setup_vim/neovim/scripts/install_dependencies_root.sh && \
    ./setup_vim/neovim/scripts/install_dependencies_user.sh

git clone $REPO_URL
cd ae_tutorial
cp .env.sample .env
# Specify your environments
vi .env
docker compose build
docker compose up -d
docker compose exec app sh
./train.sh
docker compose down