GODIVA: Generating Open-DomaIn Videos from nAtural Descriptions
This work addresses the challenge of text-to-video generation for broader applications, though it is incremental as it builds on existing methods with a focus on open-domain generalization.
The authors tackled the problem of generating videos from text by proposing GODIVA, an open-domain pretrained model that uses a three-dimensional sparse attention mechanism and is trained on a large-scale dataset of over 136 million text-video pairs, achieving good zero-shot capability on unseen texts.
Generating videos from text is a challenging task due to its high computational requirements for training and infinite possible answers for evaluation. Existing works typically experiment on simple or small datasets, where the generalization ability is quite limited. In this work, we propose GODIVA, an open-domain text-to-video pretrained model that can generate videos from text in an auto-regressive manner using a three-dimensional sparse attention mechanism. We pretrain our model on Howto100M, a large-scale text-video dataset that contains more than 136 million text-video pairs. Experiments show that GODIVA not only can be fine-tuned on downstream video generation tasks, but also has a good zero-shot capability on unseen texts. We also propose a new metric called Relative Matching (RM) to automatically evaluate the video generation quality. Several challenges are listed and discussed as future work.