In this project, I have designed and trained a model to take natural language captions in english and generate images relevant to the image captions.
I have used skip-thoughts to encode the input caption before feeding to the generator and to the discriminator.
The method used is inspired by Generative Adversarial Text to Image Synthesis.
In order to clone and reproduce results from this repository, you will need to install:
- Python 2.7
- Numpy
- Tensorflow
- Theano (For using skip-thoughts)
- OpenCV
Feel free to clone and add extensions. Pull Requests are welcome.
- Vijay Veerabadran - Vijay V
Included below are a few samples from our experiments:
| Input caption | Generated image |
|---|---|
| these flowers have an open face with many pale pink petals. |  |
| this flower is red in color, and has petals that are closely wrapped around the center. |  |
| this is a very cool flower with very bold white on the petals and a very unique middle. |  |
| multiple layers of reddish-yellow petals that decrease in size as the are closer to the top of the flower. |  |
| petals are light purple in color with longer stamens and a prominent green pistil. |  |