Conceptual

Flash-Cued Latent Diffusion Separation for Single-Image Reflection Removal

A two-stage framework that separates transmitted and reflected light from a single, possibly misaligned flash/no-flash image pair. A dual-branch latent-diffusion model conditioned on the encoded flash and no-flash latents splits the scene into transmission and reflection latents in latent space (making it robust to misalignment), and a cross-latent decoding stage conditioned on the original images restores high-resolution detail to each layer.