FVD: A new Metric for Video Generation
By Thomas Unterthiner, Sjoerd van Steenkiste, Karol Kurach, Raphaël Marinier, Marcin Michalski, Sylvain Gelly. DGS@ICLR, 2019
601 citations recorded by the source corpus.
View Paper: https://www.semanticscholar.org/paper/d4dbdeb772105a7ee780543483c6142743b20298
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
- Yoshua Bengio
- Sergey Levine
- Luc Van Gool
- Trevor Darrell
- Dacheng Tao
- Ruslan Salakhutdinov
- Abhinav Kumar Gupta
- Zhe Gan
Citations in the corpus
- VBench: Comprehensive Benchmark Suite for Video Generative Models (2023, background): “…Networks (GANs) [9, 25, 26, 30, 48, 52–55, 74], vector quantized ( However, existing metrics for video generation such as Inception Score (IS) [87], Fr´echet inception distance (FID) [37], Fr´echet Video Distance (FVD) [95, 96], and CLIPSIM [83] are inconsistent with human judgement [20, 79].”
- Imagen Video: High Definition Video Generation with Diffusion Models (2022, methodology): “We report FID evaluated on the first upsampled frame; FVD score is excessively noisy for the (cid:15) -prediction model.”
- MagicVideo: Efficient Video Generation With Latent Diffusion Models (2022, methodology): “We can observe the directed attention can substantially reduce the FVD.”
- Genie: Generative Interactive Environments (2024, methodology): “For video fidelity we use the Frechet Video Distance (FVD), a video-level metric, which has been shown to have a high level of alignment to human evaluation on video quality (Unterthiner et al., 2019).”
- DynamiCrafter: Animating Open-domain Images with Video Diffusion Priors (2023, methodology): “…the quality and temporal coherence of synthesized videos in both the spatial and temporal domains, we report Fr ´ echet Video Distance (FVD) [72] as well as Kernel Video Distance (KVD) [72], which evaluate video quality by measuring the feature-level similarity between synthesized and real…”
- Pros and Cons of GAN Evaluation Measures: New Developments (2021, methodology): “FVD considers a distribution over videos, thereby avoiding the drawbacks of frame-level measures.”