Dance-conditioned Artistic Music Generation by Creative-GAN

We present a novel adversarial, end-to-end framework based on Creative-GAN to generate artistic music conditioned dance videos. Our proposed takes the visual and motion posture data as input, then adopts quantized vector audio representation complex corresponding input. However, GAN algorithm just imitate reproduce works what humans have created, instead of generating something new creative. Therefore, we newly introduce Creative-GAN, which extends original two discriminators, one is determine whether it real music, other classify style. The paper shows that our can interesting not found in training dataset. To evaluate model, comprehensive evaluation scheme introduced make subjective objective evaluation. Compared with advanced methods, experimental results performs better measureing rhythm, generation diversity, dance-music correlation overall quality generated music.