- Original Pdf: pdf
- Data: [BAIR Robot Pushing](https://paperswithcode.com/dataset/bair-robot-pushing), [Kinetics](https://paperswithcode.com/dataset/kinetics), [Kinetics-600](https://paperswithcode.com/dataset/kinetics-600), [Moving MNIST](https://paperswithcode.com/dataset/moving-mnist)
- Keywords: autoregressive models, video prediction, generative models, video generation
- TL;DR: We present a novel autoregressive video generation that achieves strong results on popular datasets and produces encouraging continuations of real world videos.
- Abstract: Due to the statistical complexity of video, the high degree of inherent stochasticity, and the sheer amount of data, generating natural video remains a challenging task. State-of-the-art video generation models attempt to address these issues by combining sometimes complex, often video-specific neural network architectures, latent variable models, adversarial training and a range of other methods. Despite their often high complexity, these approaches still fall short of generating high quality video continuations outside of narrow domains and often struggle with fidelity. In contrast, we show that conceptually simple, autoregressive video generation models based on a three-dimensional self-attention mechanism achieve highly competitive results across multiple metrics on popular benchmark datasets for which they produce continuations of high fidelity and realism. Furthermore, we find that our models are capable of producing diverse and surprisingly realistic continuations on a subset of videos from Kinetics, a large scale action recognition dataset comprised of YouTube videos exhibiting phenomena such as camera movement, complex object interactions and diverse human movement. To our knowledge, this is the first promising application of video-generation models to videos of this complexity.