Efficient All-in-One Image Restoration With Adaptive Frequency Enhancement

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• All-in-one image restoration aims to address multiple degradation types in a single framework.
• Existing methods often suffer from high computational costs due to complex auxiliary branches.
• Efficient and effective restoration for diverse image degradations remains a challenge.

Purpose of the Study:

• To propose an efficient all-in-one image restoration network, AdaptIR.
• To introduce an Adaptive Frequency Enhancement Module (AFEM) for improved restoration.
• To achieve state-of-the-art performance with high computational efficiency.

Main Methods:

• Developed AdaptIR, an efficient network with adaptive frequency enhancement.
• Designed AFEM to couple frequency learning with adaptive convolutions for spatially adaptive restoration.
• Introduced a lightweight backbone with a Receptive Field Expansion Module (RFEM) using wavelet transform coefficients.

Main Results:

• AdaptIR achieves state-of-the-art performance on all-in-one image restoration tasks with multiple degradations.
• The model demonstrates high computational efficiency compared to existing methods.
• AFEM effectively learns pixel-wise attention weights for frequency-aware restoration.

Conclusions:

• AdaptIR provides an efficient and effective solution for all-in-one image restoration.
• The proposed AFEM and RFEM modules contribute to improved restoration quality and efficiency.
• AdaptIR is versatile and can be extended to single-degradation tasks and domain-specific applications.