FMA-Net++: Motion- and Exposure-Aware Real-World Joint Video Super-Resolution and Deblurring
Geunhyuk Youk, Jihyong Oh, Munchurl Kim
2025-12-05
Summary
This paper introduces a new method, FMA-Net++, for improving the quality of videos that are blurry and have inconsistent brightness, which often happens when videos are taken while moving or in low light.
What's the problem?
Existing video restoration techniques often struggle with videos that have both motion blur and varying levels of brightness across different parts of the video or even within the same shot. This is because cameras often adjust brightness automatically, leading to these issues, and previous methods didn't really focus on fixing both problems at the same time. It's a common problem with videos from phones or cameras trying to capture good images in difficult conditions.
What's the solution?
FMA-Net++ tackles this by looking at the entire video sequence at once, instead of frame by frame. It uses a special building block that refines the video quality while also considering how the brightness changes in each frame. Crucially, it figures out how the blur and brightness issues are related to both movement and exposure settings, and then uses that information to restore the video. The system learns what causes the problems first, then uses that knowledge to fix them, making it more accurate and faster.
Why it matters?
This work is important because it provides a more realistic way to restore videos that are commonly found in the real world. The new method performs better than previous techniques, not only in terms of visual quality but also in how quickly it can process videos, and it works well even on videos it hasn't seen before. They even created new, more challenging test videos to prove its effectiveness.
Abstract
Real-world video restoration is plagued by complex degradations from motion coupled with dynamically varying exposure - a key challenge largely overlooked by prior works and a common artifact of auto-exposure or low-light capture. We present FMA-Net++, a framework for joint video super-resolution and deblurring that explicitly models this coupled effect of motion and dynamically varying exposure. FMA-Net++ adopts a sequence-level architecture built from Hierarchical Refinement with Bidirectional Propagation blocks, enabling parallel, long-range temporal modeling. Within each block, an Exposure Time-aware Modulation layer conditions features on per-frame exposure, which in turn drives an exposure-aware Flow-Guided Dynamic Filtering module to infer motion- and exposure-aware degradation kernels. FMA-Net++ decouples degradation learning from restoration: the former predicts exposure- and motion-aware priors to guide the latter, improving both accuracy and efficiency. To evaluate under realistic capture conditions, we introduce REDS-ME (multi-exposure) and REDS-RE (random-exposure) benchmarks. Trained solely on synthetic data, FMA-Net++ achieves state-of-the-art accuracy and temporal consistency on our new benchmarks and GoPro, outperforming recent methods in both restoration quality and inference speed, and generalizes well to challenging real-world videos.