VapourSynth "dedup" idea, potential script speed enhancement?

[Frankysan]

I'm currently doing an upscale encode using VapourSynth. I had an idea on a possible way to speed the process up, but lack the scripting expertise to make it work.

Example: I'm encoding animated content, with duplicate identical frames now and then. But it's not in a consistent pattern, so a simple decimate wouldn't work without sacrificing frames.

The idea is to make a comparison on every new frame to the frame previous. If the differences are above a set threshold, the new frame gets processed as normal. But, if it's below the threshold, the rest of the processing for that frame gets skipped, and instead the processed previous frame gets presented instead. My thought is that this could speed up processing if the script is doing a lot of work, like in my case with AI upscaling.

Does anyone think this might be feasible?

[Selur]

In theory, the idea should work if the resize method only works on one frame at a time and doesn't take other frames into account.
You probably would have to write something similar to FillDuplicateFrames.

Cu Selur

[lollo]

In AviSynth you could write a function to detect the difference betwen 2 frames using RT_FrameDifference (http://avisynth.nl/index.php/RT_Stats), then if the difference is lower than a threshold, you identify the frame that does not need to be processed. You can skip these frames using FrameSel (http://avisynth.nl/index.php/FrameSel) for further processing.

Here a procedure developed to identify "bad" fields/frames in an analog capture, can be used as an example: https://forum.doom9.org/showthread.php?t=183582

In VapourSynth it should be similar.

[jagabo]

DeDup() is already set up to do that in AviSynth. The problem is the resulting video (or frames if you're saving image sequences) is variable frame rate. So you have to keep track of the timecodes for each frame. DeDup produces a timecode file for you but later processing has to be able to use it.