ReflexSplit: Single Image Reflection Separation via Layer Fusion-Separation

TL;DR

ReflexSplit tackles the challenging problem of Single Image Reflection Separation under nonlinear mixing by introducing a dual-stream architecture with explicit layer fusion-separation. The Cross-scale Gated Fusion (CrGF) enables adaptive, multi-scale feature coordination across decoder depths, while Layer Fusion-Separation Blocks (LFSB) enforce layer-specific disentanglement through early fusion and differential attention. A curriculum training strategy progressively strengthens differential separation, resulting in stable optimization and improved perceptual quality. Extensive experiments on synthetic and real benchmarks demonstrate state-of-the-art performance with robust generalization and competitive efficiency, underscoring the method's practical impact for real-world reflection suppression and texture preservation.

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.

Figures (17)

• Figure 1: Challenging examples on OpenRR-1K openrr1k. Both DSIT hu2024single and RDNet zhao2024reversible exhibit transmission-reflection confusion with incomplete reflection separation, leading to annoying artifacts or details distortion. ReflexSplit achieves better separation through explicit fusion-separation and multi-scale coordination.
• Figure 2: Layer-wise feature disentanglement comparison. DSIT hu2024single suffers from progressive transmission-reflection confusion in deep layers, exhibiting blurred boundaries and residual transmission leakage. Our method maintains clear layer distinction across all depths, preventing feature entanglement and achieving consistent reflection suppression.
• Figure 3: Overview of ReflexSplit. Dual-branch encoders extract hierarchical features: GFEB (Swin Transformer) for global semantics $\{\mathbf{P}_\ell\}$ and LFEB (MuGI-based) for local textures $\{\mathbf{E}_\ell\}$. CrGF (Section \ref{['sec:crgf']}) adaptively aggregates these multi-scale features across decoder depths, while LFSB (Section \ref{['sec:lfsb']}) alternates between fusion (cross-stream complementarity) and differential separation (layer disentanglement) to progressively refine dual streams. Curriculum training (Section \ref{['sec:training']}) progressively strengthens separation via depth-dependent initialization and epoch-wise warmup. Outputs: $\mathbf{T}$, $\mathbf{R}$, and residual $\mathbf{RR}$, which captures nonlinear interactions.
• Figure 4: Complementary roles of MuGI and proposed CrGF. (a) MuGI hu2023single focuses on dual-stream feature interaction: transmission and reflection features exchange information via channel-wise gating at each decoder level. (b) CrGF focuses on multi-scale feature integration: adaptively aggregating hierarchical encoder features and decoder context across scales to maintain gradient stability and feature consistency throughout decoding.
• Figure 5: Layer Fusion-Separation Block (LFSB). LFSB alternates between fusion (shared structure) and separation (layer disentanglement): (1) Bidirectional projection aligns transmission-reflection features; (2) Dual-dimensional attention (SA + CA) models spatial and inter-layer dependencies; (3) Differential operators $\mathbf{A}^{t} - \lambda_\ell \mathbf{A}^{r}$ suppress cross-stream interference; (4) FFNs with residuals integrate separated features.