n00mkrad / flowframes

Flowframes Windows GUI for video interpolation using DAIN (NCNN) or RIFE (CUDA/NCNN)
GNU General Public License v3.0
1.53k stars 117 forks source link

Enhancement Ideas: StaxRip + Waifu2x #311

Open tidypy opened 7 months ago

tidypy commented 7 months ago

Pre-Process Features:

(intended to be ultrafast) most of these listed below are realtime or faster, and utilized native legacy hardware. Intended to make improvements, cleanup, and prepare video.
(they really do compete or outperform AI in many business requirements).

Reason for Request:
I will Pre-Process (AVsynth, FF, Shaders), then use FlowFrame or some NN, then I must Post-Process again (StaxRip). --> This is a waste.
The reason for this effort is to, improve slightly, then play, without the need for huge processing time or resources.
Example: My netflix series that I Cached pre-flight. Abismal quality that I pay for; So I will enhance, clean, 10-bit encode, and Export.

Now I have smaller files with Satisfactory higher quality, that are: color coded, 1440 60p episodes. It should Not Take, 1hr per minute or 1.5fps with AI. This is a tragedy for a mere 3090 aficionado. (hopefully in the future a satisfactory RipUP should take .5x the runtime on an ARM archyPI.)

List of Shaders I use: glsl Shaders Anime4K_Clamp_Highlights.glsl; Anime4K_Restore_CNN_VL.glsl; Anime4K_Upscale_CNN_x2_UL.gls /Anime4K_AutoDownscalePre_x4.glsl; Anime4K_Upscale_CNN_x2_S.glsl Anime4K_Thin_VeryFast.glsl; Anime4K_Thin_HQ.glsl FSRCNN_x2_r2_32-0- Anime4K_Denoise_Bilateral_Median.gls Anime4K_Upscale_DTD Anime4K_Upscale_Denoise_CNN_x2_UL.gls Anime4K_Deblur_DoG Anime4K_Deblur_DoG_X2.glsl FSRCNNX_x2_8-0-4-1_LineArt.gls nnedi3-nns256-win8x6.hoo RAVU-lite-ar-r4.hook nnedi3-nns256-win8x4.hook ravu-lite-r4.hook; FSRAMD-FidelityFX-Super-Reso-v1-mpv.glsl;