Running Inference without GT

[Saartjes]

Your work looks amazing! I want to try to run this on a dataset of my own to improve the quality of a video. As this video is blurry of it’s own I do not have a non-blurry ground truth available. Is there any way to run the model without?

[csbhr]

You can copy the burry video and take it as the placeholder of GT. Then you can run the inference code.

In addition, if you want to get the whole deblurred video, you can copy the border frames to make this code consider border frames.

For example:

burry frames [1, 2, 3, 4, 5] -> deblurred frames [3]

copy the border frames:
burry frames [1, 1, 1, 2, 3, 4, 5, 5, 5] -> deblurred frames [1, 2, 3, 4, 5]

[Etienne66]

This is the same as issue #12, which I fixed in my fork.

As for the example. I usually do it a bit different to keep motion in the frames even if it is in the reverse direction.
blurry frames [3, 2, 1, 2, 3, 4, 5, 4, 3] -> deblurred frames [1, 2, 3, 4, 5]

I'm going to try to figure out how to do this automatically in inference.py but I haven't gotten around to it yet.