Awesome World Model Hub Papers · Datasets · Projects
← Back to datasets

GODIVA

2021 Datasets
Preview for GODIVA

Dataset Analysis

GODIVA is an open-domain text-to-video pretrained model using auto-regressive generation with 3D sparse attention, pretrained on Howto100M and evaluated with a new Relative Matching metric.

Provenance

Collected from papers.

Derived from paper: GODIVA: Generating Open-DomaIn Videos from nAtural Descriptions

Resources