Tutorial for generative models and Meta-Learning

This was originally produced as a tutorial of generative models and Meta-Learning(MAML) for Samsung Advanced Institute of Technology. There are some holes in the code to fill. See the powerpoint material for details.

How to run

• Download dataset

~/se_tutorials$ sh download_dataset.sh

• Create an anaconda environment

~/se_tutorials$ conda create -n (env name)
~/se_tutorials$ conda activate (env name)
(env name) ~/se_tutorials$ conda install --file requirements.txt

• Run tutorials

(env name) ~/se_tutorials/(topic name)$ sh command.sh

• topic names are: vae, dcgan, pix2pix, cyclegan, srgan, maml
• Edit command.sh in each topic directory for different settings. GPU number, flag changes etc..

References

• vae: https://github.com/hwalsuklee/tensorflow-mnist-VAE
• dcgan: https://gist.github.com/NeuroWhAI/d852119473cdadce066503af4196bc1a
• pix2pix: https://github.com/yenchenlin/pix2pix-tensorflow
• cyclegan: https://github.com/xhujoy/CycleGAN-tensorflow
• srgan: https://github.com/trevor-m/tensorflow-SRGAN
• meta-learning: https://github.com/cbfinn/maml