On the Content Bias in Fréchet Video Distance

arXiv 2024 25.8 benchmark

TLDR

FVD metric is biased toward per-frame quality over temporal realism; self-supervised video features reduce this bias.

Reasoning

The paper provides empirical evidence of FVD's temporal insensitivity and traces the bias to supervised video classifier features. Its main weakness is that the analysis is limited to FVD and does not address broader video generation or world model evaluation.

Read-first score

Read-first score 25.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 0.

Recency 8%

75.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2024

Methodology quality 25%

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

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%

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: 65.

Keyword Scores

world model

world simulator

generative world model

interactive world model

video world model

world dynamics prediction

model-based reinforcement learning world model

Resources

Paper