GLOW-FM enhances low-light images with one latent flow step
Original title: GLOW-FM: Efficient Generative Low-Light Image Enhancement with One-Step Latent Flow Matching
The small variant uses a frozen Tiny-VAE and Terminal Velocity Matching at 34 GFLOPs and one function evaluation.
GLOW-FM is a generative low-light image enhancement method that performs conditional transport in the latent space of a frozen Tiny-VAE. Terminal Velocity Matching is used so enhancement needs a single function evaluation, which the authors say is its first use for image restoration and LLIE. On real-world benchmarks they report competitive quality and state-of-the-art cross-dataset generalization. Code is planned for release when the paper is published.
Unconfirmed: Quality and generalization claims are the authors' experimental results; code is not yet public.
- Matched one-step comparisons are cited as support for choosing TVM.
- The small variant is specified at 34 GFLOPs.
- Inference is one function evaluation rather than iterative diffusion sampling.
- Public code is not available until publication, according to the paper.
From the original article
1 min read at huggingface.co
Low-light image enhancement (LLIE) aims to recover well-exposed and perceptually faithful images from severely degraded observations. Recent diffusion-based generative methods achieve strong perceptual quality by leveraging powerful image priors, but their iterative sampling incurs substantial computational cost, limiting practical deployment.
By Ariel Lapid, Yuval Haitman, Reuven Peretz · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.