video_to_zsm.py issue

[DaehanKim-Korea]

Running the file using pretrain(weights) does nothing after the following output.

The required ffmpeg has been installed. In any case, I think the installation method may be wrong. What installation method did the author use? Or can you share the docker image?

[MrTornado24]

I met the same question

[Mukosame]

Sorry for the late reply!
It's probably a problem with the ffmpeg installed in your system. I encountered such issue too.
I will fix it later by changing the ffmpeg command to cv2. Before that, you can inference on video using a two-step way:

• extract them into frames with this function: https://github.com/Mukosame/Zooming-Slow-Mo-CVPR-2020/blob/master/codes/utils/util.py#L324
• run python test.py for the extracted frames
• convert the frames back to video using this function: https://github.com/Mukosame/Zooming-Slow-Mo-CVPR-2020/blob/master/codes/utils/util.py#L346

Hope this solve your problem!

[EssenOH]

it is because the GPU or CPU memory ran out.
if you use GPU with DCNv2 module, it is definitely GPU memory limitation and if you wait for a long time, it will return you error message.
currently, it works with small resolution of videos with default configurations and I could test it with 0.5K resolution to generate 2K resolution.
even the 2 stage solution, it has the same issue because the problem came from the allocation to generate the temporal .hr_delme

[13610JACK]

The issue is not solved until now.