search

styler00dollar / VSGAN-tensorrt-docker

Using VapourSynth with super resolution and interpolation models and speeding them up with TensorRT.
BSD 3-Clause "New" or "Revised" License
288 stars 30 forks source link

Is there anything i should be using for Anime upscale from "SIMILARITY" & "SCENE DETECT" ? (Question/Help) #24

Closed Anon1337Elite closed 1 year ago

Anon1337Elite commented 1 year ago

Sorry if this are some shit questions, i am just trying to understand if anything would help me with this and possibly speed up the process. Thank you.

I am not familiar on how these work. Should i at least be using the one with no AI Detection ?

clip = core.misc.SCDetect(clip=clip, threshold=0.100)

Also anything to use from here that would help with anime ?

###############################################
# SIMILARITY
# Set properties in clip for it to be applied
# SSIM for deduplication in frame interpolation

# offs1 = core.std.BlankClip(clip, length=1) + clip[:-1]
# offs1 = core.std.CopyFrameProps(offs1, clip)
# 0 = PSNR, 1 = PSNR-HVS, 2 = SSIM, 3 = MS-SSIM, 4 = CIEDE2000
# clip = core.vmaf.Metric(clip, offs1, 2)
styler00dollar commented 1 year ago

There are 2 kinds of dedup, vfi dedup/sc and upscale dedup. I guess I could write it more clearly in the readme, but everything is in the config file. https://github.com/styler00dollar/VSGAN-tensorrt-docker/blob/1254e8e51233f9e849d879f897f726464d6dd13a/inference_config.py#L92-L97

Anon1337Elite commented 1 year ago

There are 2 kinds of dedup, vfi dedup/sc and upscale dedup. I guess I could write it more clearly in the readme, but everything is in the config file.

https://github.com/styler00dollar/VSGAN-tensorrt-docker/blob/1254e8e51233f9e849d879f897f726464d6dd13a/inference_config.py#L92-L97

In your opinion is it worth using, is the speed gain large or is it something that might mess up and take frames as dupes when they arent ?

styler00dollar commented 1 year ago

It depends on your content of course. Average speedup was around 30% on animated content.