LiBrA-Net: Lie-Algebraic Bilateral Affine Fields for Real-Time 4K Video Dehazing

2026-05-12

AI Summary

This work addresses the absence of evaluation benchmarks and the computational inefficiency of existing ultra-high-definition video dehazing methods, which struggle to achieve real-time 4K processing on consumer-grade GPUs. The authors propose a novel approach that formulates dehazing as a per-pixel affine transformation governed by a low-frequency depth field. This is realized through a low-resolution spatiotemporal bilateral grid predicting an affine field, augmented with an input-guided branch to recover high-frequency details. Notably, the method introduces Lie algebra regularization and Cayley parameterization to enforce invertible GL(3) transformations. The study contributes the first 4K paired video dehazing benchmark, UHV-4K, annotated with depth, transmission maps, and optical flow. The proposed model achieves state-of-the-art performance on UHV-4K, REVIDE, and HazeWorld, delivering real-time 25 FPS native 4K inference on a single GPU with only 6.12 million parameters.

Abstract

Currently, there is a gap in the field of ultra-high-definition (UHD) video dehazing due to the lack of a benchmark for evaluation. Furthermore, existing video dehazing methods cannot run on consumer-grade GPUs when processing continuous UHD sequences of 3--5 frames at a time. In this paper, we address both issues with a new benchmark and an efficient method. Our key observation is that atmospheric dehazing reduces to a per-pixel affine transform governed by the low-frequency depth field, which can be compactly encoded in bilateral grids whose prediction cost is decoupled from the output resolution. Building on this, we propose LiBrA-Net, which factorizes the spatiotemporal affine field into a spatial--color and a temporal bilateral sub-grid predicted at a fixed low resolution, fuses their coefficients in the $\mathfrak{gl}(3)$ Lie algebra under group-theoretic regularization, maps the result to invertible GL(3) transforms via a Cayley parameterization, and restores high-frequency detail through a lightweight input-guided branch. We further release UHV-4K, the first paired 4K video dehazing benchmark with depth, transmission, and optical-flow annotations on every frame. Across UHV-4K, REVIDE, and HazeWorld, LiBrA-Net sets a new state of the art among compared video dehazing methods while running native 4K at 25 FPS on a single GPU with only 6.12 M parameters. Code and data are available at https://anonymous.4open.science/r/LiBrA-Net-42B8.

Problem

• video dehazing
• ultra-high-definition
• real-time processing
• benchmark dataset
• consumer-grade GPU

Innovation

• Lie algebra
• bilateral grids
• affine fields
• real-time 4K dehazing
• video dehazing benchmark