deepmagd / shenaniGAN

Work in progress
MIT License
1 stars 0 forks source link

Feature: Data pipelines for captioned image input implemented #4

Closed alecokas closed 4 years ago

alecokas commented 4 years ago

Set up data processing and training for Text-to-Image alongside Image-to-Image training

You can now run: python main.py test -d birds-with-text

and

python main.py test -d birds and the data will be automatically downloaded, formatted and loaded ready for training.

A trainer from Text-to-Image is set up and ready to be used. The image-to-image trainer is working as is.

This code involves some refactoring as well.

NOTE: This is only for the pretrained caption embeddings. When we want to use our own embeddings, further work will be required.