Singularity42 / Sync-DRAW

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

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

Tensorflow implementation for the paper Sync-DRAW: Automatic GIF Generation using Deep Recurrent Attentive Architectures by Gaurav Mittal*,Tanya Marwah* and Vineeth N. Balasubramanian accepted as an oral paper at ACM-Multimedia 2017 (*Equal Contribution).

Sync-DRAW architecture. (a) The read-write mechanisms and the R-VAE; (b) Conditioning the R-VAE's decoder on text features extracted from captions for text-to-video generation; (c) Canvases generated by Sync-DRAW over time (each row is a video at a particular time step, last row is the final generated video).

Results

One Digit

Without Caption	With Caption
	Digit 5 is moving left and right.

Two Digit

Without Caption	With Caption
	digit 2 is going up and down and digit 0 is going left and right.

KTH Dataset