Sync-DRAW: Automatic GIF Generation using Deep Recurrent Attentive Architectures

Mittal, G and Marwah, T and Balasubramanian, Vineeth N (2016) Sync-DRAW: Automatic GIF Generation using Deep Recurrent Attentive Architectures. arXiv. pp. 1-9.

Preview

Text (arXiv copy)
1611.10314.pdf - Accepted Version
Download (1MB) | Preview

Abstract

This paper introduces a novel approach for generating GIFs called Synchronized Deep Recurrent Attentive Writer (Sync-DRAW). Sync-DRAW employs a Recurrent Variational Autoencoder (R-VAE) and an attention mechanism in a hierarchical manner to create a temporally dependent sequence of frames that are gradually formed over time. The attention mechanism in Sync-DRAW attends to each individual frame of the GIF in sychronization, while the R-VAE learns a latent distribution for the entire GIF at the global level. We studied the performance of our Sync-DRAW network on the Bouncing MNIST GIFs Dataset and also, the newly available TGIF dataset. Experiments have suggested that Sync-DRAW is efficient in learning the spatial and temporal information of the GIFs and generates frames where objects have high structural integrity. Moreover, we also demonstrate that Sync-DRAW can be extended to even generate GIFs automatically given just text captions.

[error in script]

IITH Creators: