Open semjon00 opened 1 year ago
Batching from directory loads all the images at once. Not ideal.
its very not ideal, because it can hit C runtime max open files limit
And when you have just 1 more file than 8192, it throws OSError: [Errno 24] Too many open files: './models/leres/res101.pth'
error
Which is very annoying when trying to do a whole video, since you have to convert it in like 4 rus...
@T0biasCZe Please try out the new Single Video mode :) By the way, 8192 is a lot of files O_O This can be patched, I will look into it in the following week or two.
By the way, 8192 is a lot of files O_O
Wellllll, at 24fps thats just 5 minutes 41seconds :D and considering it runs at 1.5files/s (with res101) it doesnt take that long
Also, with the video mode, if the conversion stops mid generation (for example error or something), the already generated images will propably be lost and it will be uncontinuable (which is issue for 20 minute vids...) Unless it could be made to output the converted frames into jpegs instead of back into another video π€And then when converting the frames, it would check if the frame already exists in the folder and potencially skip it π€
Yeah like expected, mid conversion threw error, video gon π
Unless it could be made to output the converted frames into jpegs instead of back into another video thinkingAnd then when converting the frames, it would check if the frame already exists in the folder and potencially skip it thinking
This would make it really complicated. I will think about this, maybe there will be a solution in the future.
@T0biasCZe This is not a memory issue (the video was done except saving video from ready frames). ~Please ensure that the plugin is properly installed. For whatever reason backbone.py was not properly updated. How did you install the update?~ Ooh! Found an error! Thanks!
@T0biasCZe Please update the extension and try again - please let me know if everything works now.
the video was done except saving video from ready frames
so it does save into jpegs first? π
No, no. I meant that the images that the video would consist of were already created - 16-bit grayscale png-s for depthmaps and/or 24-bit images for colorvideos. These images are stored in RAM and not saved. What was left is to bundle them into a video. Please try again (after updating the extension), it should just work.
Jpeg/Png doesnt matter. And i meant the already processed and finished images, not the raw images from the input video...
sooo... It should be possible to make the processing continue in case it for any reasons stopped during the conversion π
And yeah the new update forks.... (but that wont help with other mid process interuptions)
Please send the stacktrace... You know, otherwise I can't help.
sooo... It should be possible to make the processing continue in case it for any reasons stopped during the conversion upside_down_face
Yep, a good idea, but not a priority right now. There are some more pressing issues at the moment... If you know how to fix it, you could create a MR, I would happily review and merge :+1:
Please send the stacktrace... You know, otherwise I can't help.
i didnt mean any specific error right now, but I meant overall if it stopped later, for example system running out of ram ehm memory laks, windows crashing, etc. that would make the conversion get interrupted mid run
@T0biasCZe Sorry, misread your message
Pls do output files one at a time, not all at the very end,,, so when there is an error I do not have to start all over, I can continue with the files that aare not processed... it was working like that before... I am having problems with large files...
Known VRAM leaks:
Known RAM leaks: