Hsu, Cheng-WeiCheng-WeiHsuLee, Ming-SuiMing-SuiLee2026-04-242026-04-242025-11-28[9798331572068]https://www.scopus.com/record/display.uri?eid=2-s2.0-105030498641&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/737504Reflections are ubiquitous in our daily lives, making it inevitable for standard photographic equipment to capture unwanted reflected objects when taking images. Reflections degrade image aesthetics and can impair downstream vision tasks, making their removal from a single image a long-standing research focus. Although deep learning-based methods have made significant progress compared to traditional approaches in recent years, their performance is still limited by two fundamental issues: overly simplified reflection model assumptions and the domain gap between synthetic and real-world reflection images. A diffusion model-based approach is introduced to reduce dependency on assumptions, using a dual-stream network architecture to simultaneously predict residuals and reflection layers, thereby enhancing the diffusion model's ability to capture complex data distributions. Additionally, we employ physically based rendering techniques to generate the necessary training datasets, narrowing the gap between real-world images and synthetic data. Experimental results on the benchmark data demonstrate that the proposed model achieves performance comparable to state-of-the-art methods.falseconference paper10.1109/apsipaasc65261.2025.112491922-s2.0-105030498641