Video Stability Evaluation Metrics

[jorie-peng]

Thanks for your great job!
I saw you used evaluation metrics in the paper but would it be possible to share the implementation of the stability metrics you used? Unfortunately, I haven't found a good implementation of the stability, and accumulated optical flow metrics which is publicly available yet.

[luomomo]

Hi，
I am testing the metric in the code with some simple simulation video
#Crop(rotation 10deg) and scale the origin video
But I can not get a good result. The fov is always 1.0
#using static video(not including any motion)
the stability score is not 1.0
Does anyone meet the same issue?

[zhmeishi]

For the stability score in the metrics.py we reuse it from https://github.com/jinsc37/DIFRINT/blob/master/metrics.py

We didn't do the test as you did but I believe the score should be close to 1 in your setting.

Here are some hints I have. In the file https://github.com/googleinterns/deep-stabilization/blob/master/dvs/metrics.py
In line 187, we used SIFT to map two successive frames. You may test whether the sourcePoints and destinationPoints are the same.
If they are the same, in line 203, you may find whether the homography matrix is an identity matrix.
In lines 233-237, we do a fast Fourier transform. In lines 247-248, you may test whether the first five entries keep all the information.

[luomomo]

O(∩_∩)O非常感谢您给我发来邮件，稍后我会仔细阅读。