'frames' folder split leaving random frame out, breaking post-upscale encode step

[k4yt3x]

'frames' folder split leaving random frame out, breaking post-upscale encode step

[k4yt3x]

@cr08 I'll will take a look at this problem.
The multi-threading function has just been added not too long ago, and this might as well be a fatal bug.

As for the performance, here's what I got on my system.

• My 16 DDR4 is used up to 95% when upscaler threads are initializing, and soon it drops down to normal, so does the CPU usage drop down quite a bit.
• The GPU computing usage gets up to 90-100% for the entire duration of the upscaling, while the VRAM is only half-filled.

[k4yt3x]

Also, make sure you're using the newest version of video2x. The latest changes has been made for fixing bugs within the multithreading function.

Also, the part where video2x places frames into folders it at line 150, video2x.py. Maybe you can find some issues there?

[cr08]

I'm running the latest commit. Downloaded fresh this morning.

I wonder if it might be a simple math issue. I rechecked and the exact number of frames in this video is 35,763. 3 threads means it is divided up to 11,891 frames per thread with one frame left over.

[k4yt3x]

@cr08 I think that's exactly where the problem is.

I'm running Kali right now and doesn't have access to a windows workstation at the moment. I'll make the changes when I get back home.

Or you can submit a pull request (wink :P

[k4yt3x]

@cr08 Actually, never mind. I was able to RDP into my server and test the patch.
I have just pushed a new commit into the repo which should fix the problem.

This might be a less stupid way of distributing images evenly.

    # Evenly distribute images into each folder
    # until there is none left in the folder
    for image in frames:
        # Move image
        shutil.move(image, thread_folders[0])
        # Rotate list
        thread_folders = thread_folders[-1:] + thread_folders[:-1]

[cr08]

So far so good. Ran a few shorter clips through with varying frame counts and no issues.

[k4yt3x]

If all's fine, then I'll close this issue.
We can reopen it any time if it's still malfunctioning.