Challenging Optical Flow Dataset and Benchmark
Si Lu, Portland State University
As a part of our work in "High-speed Video from Asynchronous Camera Array", we release an optical flow evaluation dataset that we used in our paper. In this dataset, we provide 4 video sequences that are challenging for optical flow estimation, including scenes with large camera motion, blurry fast moving object (car, train) and complicated object motions such as walking. Each of the four sequences contains about 50 input frames and corresonding interpolated ground truth frames with an interpolation factor of 0.5 (which means the interpolated frames are temporally located at the middle of two consecutive input frames). Please see the table and figures below to get an overview of each sequences.
When using this dataset in your research, we will be happy if you cite us:
Si Lu. High-speed Video from Asynchronous Camera Array.  IEEE WACV 2019. PDF, Poster, Code

    title={High-speed Video from Asynchronous Camera Array},
    author={Lu, Si},
    booktitle={2019 IEEE Winter Conference on Applications of Computer Vision (WACV)},
Related Datasets

Middlebury Stereo Evaluation: The classic stereo evaluation benchmark, featuring four test images in version 2 of the benchmark