The codes of paper "Long Text Generation via Adversarial Training with Leaked Information" on AAAI 2018. Text generation using GAN and Hierarchical Reinforcement Learning.
I think it's another kind of conditional text generation, likes translation(given text, generate translation), the start token or start hidden_state like condition. This is my guess, welcome to discuss.
In Main.py, line 22 (https://github.com/CR-Gjx/LeakGAN/blob/master/Image%20COCO/Main.py#L22 ) START_TOKEN is set to 0. I wonder is it possible to use LeakGAN to generate text samples in batch where each sample starts with a different START_TOKEN?