GAN model trained in Tensorflow and then ported to browsers using GPU-accelarated framework deeplearn.js which is now the core of tensorflow.js
You can repeat the whole experiment in 3 steps:
- Run train.py to get ./weights folder
- Build generator.ts
- Run node server and open index.html
- You are awesome!
- train.py - creates and trains the model, saves generator weights
- GAN.py - model definition (originated from this repo)
- yellowfin.py - powerful custom TF optimizer, original repo. Not really required, just a nice thing.
- ops.py and utils.py - simplified TF operations and other utility functions
- generator.ts - downloads weights and defines generator model
- index.html - simple wrapper
- other configs