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

arXiv 2016 25.9 method

TLDR

Sync-DRAW combines VAE with recurrent attention for automatic video generation, including text-to-video, evaluated on Bouncing MNIST, KTH, and UCF-101.

Reasoning

The paper presents a novel architecture and claims to be the first text-to-video generation approach, with experiments on multiple datasets. However, the abstract lacks quantitative results, comparisons, and explicit connections to world models, limiting support for those keywords.

Read-first score

Read-first score 25.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 4.

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,dataset

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=dataset,experiment

Recency 8%

Uses a gentle age decay so recent papers surface without erasing older foundations. 2016

Topical relevance 42%

5.7

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Field roles

Candidate

Rank sensitivity

Stability: volatile; rank range: 74.

Keyword Scores

video world model

world model

world dynamics prediction

world simulator

generative world model

interactive world model

model-based reinforcement learning world model

Resources

Paper