ReflexSplit: Single Image Reflection Separation via Layer Fusion-Separation

2026-01-24

AI Summary

This work addresses the challenge of disentangling transmission and reflection layers in a single image, particularly when nonlinear mixing causes ambiguity and deep decoders suffer from ambiguous feature fusion and insufficient multi-scale coordination. To this end, the authors propose ReflexSplit, a dual-stream framework that incorporates cross-scale gated fusion (CrGF) and a layer fusion-separation block (LFSB) featuring a cross-stream subtraction attention mechanism inspired by differential Transformers. The approach is further enhanced by a progressive curriculum training strategy leveraging depth-dependent initialization and iterative warm starts. Experimental results demonstrate that ReflexSplit achieves state-of-the-art performance on both synthetic and real-world datasets, significantly improving the perceptual quality and generalization capability of the separation results.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Search and Optimization: Non-convex Optimization

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Single Image Reflection Separation (SIRS) disentangles mixed images into transmission and reflection layers. Existing methods suffer from transmission-reflection confusion under nonlinear mixing, particularly in deep decoder layers, due to implicit fusion mechanisms and inadequate multi-scale coordination. We propose ReflexSplit, a dual-stream framework with three key innovations. (1) Cross-scale Gated Fusion (CrGF) adaptively aggregates semantic priors, texture details, and decoder context across hierarchical depths, stabilizing gradient flow and maintaining feature consistency. (2) Layer Fusion-Separation Blocks (LFSB) alternate between fusion for shared structure extraction and differential separation for layer-specific disentanglement. Inspired by Differential Transformer, we extend attention cancellation to dual-stream separation via cross-stream subtraction. (3) Curriculum training progressively strengthens differential separation through depth-dependent initialization and epoch-wise warmup. Extensive experiments on synthetic and real-world benchmarks demonstrate state-of-the-art performance with superior perceptual quality and robust generalization. Our code is available at https://github.com/wuw2135/ReflexSplit.

Problem

• Single Image Reflection Separation
• Transmission-Reflection Confusion
• Nonlinear Mixing
• Multi-scale Coordination
• Layer Disentanglement

Innovation

• Cross-scale Gated Fusion
• Layer Fusion-Separation Block
• Differential Separation
• Curriculum Training
• Single Image Reflection Separation