VIDM: Video Implicit Diffusion Models

Diffusion models have emerged as a powerful generative method for synthesizing high-quality and diverse set of images. In this paper, we propose video generation based on diffusion models, where the effects motion are modeled in an implicit condition manner, i.e. one can sample plausible motions according to latent feature frames. We improve quality generated videos by proposing multiple strategies such sampling space truncation, robustness penalty, positional group normalization. Various experiments conducted datasets consisting with different resolutions number Results show that proposed outperforms state-of-the-art adversarial network-based methods significant margin terms FVD scores well perceptible visual quality.