Dense Optical Flow
Dense optical flow estimation aims to compute the motion of every pixel in a video sequence, providing a rich representation of scene dynamics. Current research focuses on improving accuracy and robustness in challenging conditions (e.g., low light, fast motion) using techniques like self-supervised learning, neural implicit representations, and hybrid approaches combining deep learning with traditional methods. These advancements are driving progress in applications such as autonomous driving, 3D reconstruction, video synthesis, and robotic manipulation, where accurate motion understanding is crucial.
21papers
Papers
March 13, 2025
March 5, 2025
December 16, 2024
December 9, 2024
November 21, 2024
November 19, 2024
October 31, 2024
June 27, 2024
April 29, 2024
February 9, 2024
August 14, 2023