New FMA-Net++ framework tackles video super-resolution and deblurring

Researchers have developed FMA-Net++, a novel framework for joint video super-resolution and deblurring that addresses challenges posed by varying frame-wise exposure durations. The system utilizes Hierarchical Refinement with Bidirectional Aggregation blocks for efficient parallel processing and expands temporal receptive fields without recurrent bottlenecks. It introduces an Exposure Time-aware Modulation layer to condition features on exposure embeddings, enabling the prediction of motion- and exposure-aware degradation kernels. FMA-Net++ demonstrates state-of-the-art performance on new benchmarks, REDS-ME and REDS-RE, and shows strong out-of-distribution capabilities on existing datasets like GoPro. AI

IMPACT Introduces a novel approach to video restoration, potentially improving quality in applications with variable lighting conditions.

RANK_REASON Academic paper detailing a new model and benchmarks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Exposure Time-aware Feature Extractor
• Exposure Time-aware Modulation
• FMA-Net++
• Geunhyuk Youk
• GoPro
• Hierarchical Refinement with Bidirectional Aggregation

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →