OneRestore: A Universal Restoration Framework for Composite Degradation

TL;DR

This work tackles the challenge of restoring images corrupted by multiple concurrent degradations by introducing OneRestore, a transformer-based universal restoration framework that conditions restoration on scene descriptors. A Scene Descriptor-guided Transformer Block (SDTB) fuses descriptor queries with image features via cross-attention, enabling fine-grained, controllable restoration using either manual text embeddings or automatic visual attributes. The authors also develop a Composite Degradation Dataset (CDD-11) to simulate 11 degradation combinations and propose a composite degradation restoration loss that leverages negative degraded samples to enforce robust separation. Empirical results on synthetic and real-world data show OneRestore achieving state-of-the-art performance for composite degradations while maintaining competitive results on single-degradation benchmarks, with demonstrated controllability and efficiency advantages.

Abstract

In real-world scenarios, image impairments often manifest as composite degradations, presenting a complex interplay of elements such as low light, haze, rain, and snow. Despite this reality, existing restoration methods typically target isolated degradation types, thereby falling short in environments where multiple degrading factors coexist. To bridge this gap, our study proposes a versatile imaging model that consolidates four physical corruption paradigms to accurately represent complex, composite degradation scenarios. In this context, we propose OneRestore, a novel transformer-based framework designed for adaptive, controllable scene restoration. The proposed framework leverages a unique cross-attention mechanism, merging degraded scene descriptors with image features, allowing for nuanced restoration. Our model allows versatile input scene descriptors, ranging from manual text embeddings to automatic extractions based on visual attributes. Our methodology is further enhanced through a composite degradation restoration loss, using extra degraded images as negative samples to fortify model constraints. Comparative results on synthetic and real-world datasets demonstrate OneRestore as a superior solution, significantly advancing the state-of-the-art in addressing complex, composite degradations.

Figures (22)

• Figure 1: Our OneRestore allows fully controllable image restoration using scene descriptors derived from both automatic visual attribute extraction (top) and manual text embeddings (bottom).
• Figure 2: Structural comparison between One-to-One, One-to-Many, and our One-to-Composite image restoration methods.
• Figure 3: Architecture of our OneRestore. (a) Overall pipeline, where 32, 64, 128, and 256 represent the number of channels. (b) Scene descriptor generation, where scene descriptors are fed into each (c) Scene Descriptor-guided Transformer Block (SDTB) by manual text embeddings or automatic extraction based on visual attributes.
• Figure 4: Illustration of the composite degradation restoration loss.
• Figure 5: Comparison of quantitative results for different degradation scenarios on CDD-11 dataset.