Unified-Width Adaptive Dynamic Network for All-In-One Image Restoration

Understand

In contrast to traditional image restoration methods, all-in-one image restoration techniques are gaining increased attention for their ability to restore images affected by diverse and unknown corruption types and levels.

• However, contemporary all-in-one image restoration methods omit task-wise difficulties and employ the same networks to reconstruct images afflicted by diverse degradations.
• This practice leads to an underestimation of the task correlations and suboptimal allocation of computational resources.
• To elucidate task-wise complexities, we introduce a novel concept positing that intricate image degradation can be represented in terms of elementary degradation.