Diffusion Models for Video Prediction and Infilling

arXiv 2022 32.1 method

TLDR

RaMViD extends diffusion models to videos with 3D convolutions and masking for state-of-the-art video prediction and infilling.

Reasoning

The paper introduces a novel conditioning technique for video diffusion models, achieving state-of-the-art results on benchmark datasets. However, it focuses narrowly on video prediction and infilling without explicit connection to world models or reinforcement learning, limiting its scope.

Read-first score

Read-first score 32.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 9.

Recency 8%

56.5

Uses a gentle age decay so recent papers surface without erasing older foundations. 2022

Methodology quality 25%

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

Reproducibility 25%

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

Topical relevance 42%

12.9

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: 49.

Keyword Scores

world dynamics prediction

video world model

world model

world simulator

generative world model

interactive world model

model-based reinforcement learning world model

Resources

Paper