Learned Dual-View Reflection Removal

Abstract

Tianfan XueGoogle Research(a) View 1 (b) View 2 (c) Anaglyph (d) Aligned (e) Ours Figure 1: Stereo pairs (a, b) were imaged through glass and exhibit undesired reflections. The transmitted and reflective images are subject to parallax that is difficult to separate as shown in the anaglyph (c). Our reflection-invariant flow aligns the two views with respect to the transmitted image, causing all remaining parallax (in the reflection on the tissue box, for example) to be due to reflections as shown in anaglyph (d). Our synthesis network exploits this parallax to remove reflections (e).

Softmax Splatting for Video Frame Interpolation

“…Video frame interpolation is becoming more and more ubiquitous. While early techniques for frame interpolation were restricted to using block motion estimation and compensation due to performance constraints [8,20], modern graphics accelerators allow for dense motion estimation and compensation while heavily making use of neural networks [37,45,46,48]. These developments enable interesting new applications of video frame interpolation for animation inbetweening [32], video compression [63], video editing [40], motion blur synthesis [3], and many others.…”

Robust Reflection Removal with Flash-only Cues in the Wild

“…Multiple images reflection removal. Some reflection removal methods utilize the motion cue of reflection and transmission in multiple images for reflection removal [17], [18], [19], [20], [21], [22], [23]. In these motion-based methods, SIFT-flow [19], homography [17] and optical flow [20], [22], [23] are used to find correspondences among multiple images to distinguish reflection and transmission.…”

“…Some reflection removal methods utilize the motion cue of reflection and transmission in multiple images for reflection removal [17], [18], [19], [20], [21], [22], [23]. In these motion-based methods, SIFT-flow [19], homography [17] and optical flow [20], [22], [23] are used to find correspondences among multiple images to distinguish reflection and transmission. However, taking images with different motion cost more effort, and some assumptions are required (e.g., all pixels in transmission must appear in at least one image [22]).…”

Introduction to Artificial Intelligence and Robotics

“…Alternative methods have been explored using additional inputs for reflection removal. Motion-based techniques [4,12,13,27,29,33,37,46] utilize multiple images capturing different motion characteristics to separate reflections from transmissions. These methods, however, require more complex image capturing and certain assumptions, such as the visibility of all transmission pixels in at least one image [46].…”

“…"Beyond-assumption" reflections have consistently hindered previous methods from generalizing in the wild. To achieve more robust reflection removal, several works adopt multiple images or modalities to gather additional information about reflections, such as polarization images [19,23,30,34,35,49], flash images [21], and multi-view images [13,33,46]. However, these methods require additional sensors, equipment, or multiple captures, which limits their practical flexibility and scalability to diverse real-world scenarios.…”