Summary of Fld+: Data-efficient Evaluation Metric For Generative Models, by Pranav Jeevan et al.
FLD+: Data-efficient Evaluation Metric for Generative Models
by Pranav Jeevan, Neeraj Nixon, Amit Sethi
First submitted to arxiv on: 23 Nov 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG); Image and Video Processing (eess.IV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes a novel metric, Flow-based Likelihood Distance Plus (FLD+), to evaluate the quality of generated images. Unlike existing metrics like Fréchet Inception Distance (FID), FLD+ is more reliable, data-efficient, compute-efficient, and adaptable to new domains. The proposed metric uses normalizing flows, which allows for the computation of exact log-likelihood density of images from any domain. This leads to strongly monotonic behavior with respect to different types of image degradations, such as noise, occlusion, diffusion steps, and generative model size. FLD+ achieves stable results with two orders of magnitude fewer images than FID, making it a more practical choice for evaluating generated images. Additionally, the paper shows that FLD+ can be retrained on new domains, such as medical images, unlike existing metrics. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is all about finding a better way to measure how good computer-generated images are. Right now, we use something called Fréchet Inception Distance (FID) to do this, but it has some big drawbacks. The new method they propose, called Flow-based Likelihood Distance Plus (FLD+), gets around these problems by using special math techniques that allow us to measure how likely an image is to be real. This helps the metric work better with all kinds of images and even work on totally new types of pictures that we haven’t seen before. |
Keywords
» Artificial intelligence » Diffusion » Generative model » Likelihood » Log likelihood
