PWC-Net compared with Shi-Tomasi

[zyong812]

Dear Dr. Sun,

I’ve read your paper “PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume”. I am wandering how the pre-trained PWC-Net is compared to traditional method like Shi-Tomasi algorithm in terms of accuracy? I’ve read checked several papers of estimating optical with deep networks. None of them compare the results with traditional methods. I don’t know why there is no such comparison.

By the way, for a new dataset without optical flow ground truth, how should we fine-tune the pre-trained PWC-Net? Do you have any suggestions about this? Thank you so much!

Have a nice day!