st186 / BoGAN

Pytorch implementation for “Scripted Video Generation with a Bottom-up Generative Adversarial Network”

BoGAN

Pytorch implementation for “Scripted Video Generation with a Bottom-up Generative Adversarial Network”

Dataset

We have released the datasets we use in the paper “Scripted Video Generation with a Bottom-up Generative Adversarial Network”, including

• Single-Digit Bouncing MNIST GIFs (SBMG)

• Two-Digit Bouncing MNIST GIFs (TBMG)

• MSVD and MSVD Cooking

Download Text--to--Video Generation dataset:

1. Text--to--Video Generation (Baidu Yun) Password: z69q

2. Text--to--Video Generation (Google Drive)

Training

Stage I: Optimise the alignment model in DAMSM, e.g., for SBMG dataset:

bash run_one.sh

Stage II: Optimise the generator and discriminators in BoGAN, e.g., for SBGM dataset:

bash run_one.sh

Testing

bash run_test_one.sh