High dynamic range (HDR) imaging is a technique to capture a scene with a high dynamic range, which is one of the important areas in computer vision and image processing. HDR Video Reconstruction, on the other hand, is a more advanced field where the ...
High dynamic range (HDR) imaging is a technique to capture a scene with a high dynamic range, which is one of the important areas in computer vision and image processing. HDR Video Reconstruction, on the other hand, is a more advanced field where the challenge is to recover all frames of an input video as HDR, rather than just a single frame. One of the challenges in HDR Video Reconstruction is the occurrence of ghosting artifacts due to significant motion. To address this issue, some methods are proposed utilizing deep learning-based optical flow models to calculate optical flow and apply image warping to each frame for alignment before performing HDR imaging. However, previous studies lacked comprehensive evaluations of various optical flow models performance. Therefore, this paper presents an evaluation of three types of optical flow models to measure performance variations and qualitatively and quantitatively compare them.