Reversible Decoupling Network for Single Image Reflection Removal

2024-10-10

arXiv.org

AI Summary

In single-image reflection removal, existing dual-stream deep methods suffer from two key limitations: progressive loss of high-level semantic information across layers and rigid cross-stream interaction patterns. To address these, we propose the Reversible Decoupling Network (RDNet). First, we design a reversible encoder that losslessly preserves critical information during forward propagation while enabling layer-adaptive decoupling of transmission and reflection features. Second, we introduce a transmittance-aware prompt generator to dynamically calibrate dual-stream features. Third, we replace fixed interaction paradigms with an information-bottleneck-driven deep supervision architecture for adaptive feature refinement. RDNet achieves state-of-the-art performance across five mainstream benchmarks, with significant improvements in PSNR and SSIM. The source code will be made publicly available.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Algorithmic accountability and transparency on the web
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Recent deep-learning-based approaches to single-image reflection removal have shown promising advances, primarily for two reasons: 1) the utilization of recognition-pretrained features as inputs, and 2) the design of dual-stream interaction networks. However, according to the Information Bottleneck principle, high-level semantic clues tend to be compressed or discarded during layer-by-layer propagation. Additionally, interactions in dual-stream networks follow a fixed pattern across different layers, limiting overall performance. To address these limitations, we propose a novel architecture called Reversible Decoupling Network (RDNet), which employs a reversible encoder to secure valuable information while flexibly decoupling transmission- and reflection-relevant features during the forward pass. Furthermore, we customize a transmission-rate-aware prompt generator to dynamically calibrate features, further boosting performance. Extensive experiments demonstrate the superiority of RDNet over existing SOTA methods on five widely-adopted benchmark datasets. Our code will be made publicly available.

Problem

• Overcoming information loss in reflection removal networks
• Enhancing dual-stream network flexibility for better performance
• Dynamically calibrating features for improved reflection removal

Innovation

• Reversible encoder secures valuable information
• Transmission-rate-aware prompt generator calibrates features
• Flexible decoupling of transmission and reflection features