paarthneekhara/text-to-image

Combines Skip Thought Vectors for semantic caption encoding with the GAN-CLS adversarial architecture to generate 64×64 images from natural language descriptions. Built on TensorFlow and DCGAN, the implementation uses pretrained skip-thought embeddings to transform captions into fixed-length vector representations that guide the generator network. Trained on the Oxford Flowers dataset with horizontal augmentation and supports multi-image sampling per caption during inference.

2,167 stars. No commits in the last 6 months.