ravi-maurya / learning_data_aug

OpenAI Request for Research - https://blog.openai.com/requests-for-research-2/

Learning Data Augmentation

OpenAI Request for Research - https://blog.openai.com/requests-for-research-2/

Requirements

• Python 3
• Anaconda (Numpy, Matplotlib, PIL, etc)
• PyTorch
• GPU, CUDA

Tested on Ubuntu 16.04 with GeForce GTX 1080 Ti

Papers

• RenderGAN: Generating Realistic Labeled Data
• Dataset Augmentation in Feature Space
• Learning to Compose Domain-Specific Transformations
• Data Augmentation Generative Adversarial Networks
• Data Augmentation in Emotion Classification Using GANs
• Neural Augmentation and Cycle GAN
• Bayesian Data Augmentation Approach
• Learning To Model the Tail
• Tutorial on Variational Autoencoders

Approaches

1. Autoencoder: Encode --> Peturb/Interpolate --> Decode
2. Synthetic Dataset Generation with CGAN
3. Domain Adaption / Image-to-Image translation (CycleGAN)
4. Augmentation / Preprocessing Net + backprop classification loss

Posts

Autoencoder

• https://pgaleone.eu/neural-networks/2016/11/24/convolutional-autoencoders/

VAE

• http://kvfrans.com/variational-autoencoders-explained/
• https://wiseodd.github.io/techblog/2016/12/10/variational-autoencoder/
• https://wiseodd.github.io/techblog/2016/12/17/conditional-vae/

GAN

• http://blog.otoro.net/2016/04/01/generating-large-images-from-latent-vectors/

Code

Autoencoder

• https://blog.keras.io/building-autoencoders-in-keras.html

VAE

• https://github.com/kvfrans/variational-autoencoder
• https://github.com/wiseodd/generative-models/blob/master/VAE/vanilla_vae/vae_pytorch.py
• https://github.com/wiseodd/generative-models/blob/master/VAE/conditional_vae/cvae_pytorch.py

Links

• https://blog.openai.com/requests-for-research-2/
• http://rodrigob.github.io/are_we_there_yet/build/classification_datasets_results.html#4d4e495354
• https://paper.dropbox.com/doc/VAE-QXVhUneHWqtvYELdjpR9d (my notes)