Evaluation Datasets and Benchmarks for Optical Flow Algorithms: A Review

Optical flow analysis for motion estimation, image segmentation, object detection and tracking has significantly revolutionized the fields of computer vision and robotics. Optical flow datasets provide a base and benchmark for training, testing and comparison of optical flow algorithms. This paper provides a review and analysis of available datasets that can be used for training and evaluation of optical flow algorithms in various applications of computer vision and robotics. In this paper, optical flow datasets are discussed with different attributes that can be utilized for comparison, reflecting the advantages and correct usage for key implementation of a specific task. In addition to this, open research challenges for the generation of finer and preferable optical flow datasets have been discussed.