Text2image

Andrew Peng & David Lin

Data

Download the Oxford 102 dataset here and the captions here

Training

Update variable in constants.py and run train.py

TODO in order of importance

1. How to get text embeddings
   • RNN (DONE)
   • Skip thought (DONE)
   • BERT (DONE)
   • GPT (DONE)
2. Losses (DONE)
3. Get it to train (DONE)
4. Put on GCP / AWS, add GPU support (DONE)
5. Benchmarks for accuracy (DONE)
6. Loading and saving models (DONE)
7. Generate image examples (DONE)