Selur's Little Message Board

My video project is to digitize 8mm tapes and VHS tapes. I built a WindowsXP machine in order to accommodate the ATI 9800XT capture card. For 8mm tapes, I use a Sony TRV-66 camcorder which has a TBC. Posts at DigitalFAQ recommended Selur Hybrid. I have been using it for about a year. After experimenting with different denoise settings, de-halo and so on, I learned about the dev version available on request. Selur was kind enough to share it with me. 

The dev version has the AI(?) upscaler realESRGAN. I tried the different "modes" and found the ones that worked best for my 8mm and VHS tapes was realESR-general and 2X-plus. From what I gather, these upscalers work great for animated content, but for 8mm and VHS recordings, some are not appropriate.

For, for the record in case this helps others, and also to get feedback on my settings, I am sharing this.

This is my conversion workflow for VHS tapes and 8mm tapes captured with VirtualDub.

1. Use Selur Hybrid version Hybrid_dev_2023.02.05-155258 (available on request from Selur) and VapoursynthR61_torch and vsgan_models
2. These are the settings I use in Hybrid:
- Video processing: x264, mp4
- Encoding mode: Constant rate factor 1-pass, Quantization factor 18.00
- Crop/resize
- Picture crop 0,8,8,14
- Output PAR 17x15
- Resize: 1920x1440
- Filtering--Frame--Resize
- RealESRGAN
- Mode: realESR-general (or sometimes X2 works better)
- Denoise strength: 0.3

I know that a skilled video restorer can do more by tweaking the Vapoursyth settings. But after lots of experimentation, I feel like I am still a newbie and so welcome any AI that can help figure out for me the best settings.

I welcome feedback.

Please, read the 'Infos needed to fix&reproduce bugs,..'-sticky before you post about a problem.

Hi thank you forthe infos... and what about the deinterlacing? i use a blackmagic intensity pro to aquire a 625i PAL..
720x576 interlaced uncomrpessed...

i use the great QTGMC  vapoursynth plugin to obtain a 720x576 50hz deinterlaced (with VHS analog noise recorded at this resolution)
and after that i use the degrain ( Temporal Degrain ) to reduce the remaining pixels on vertical items (bars.. fences...)

after that i save using matroska (to preserve at 100% audio/video sync) Constant rate factor 1-pass quantization 14 (when using for editing better 12-13 )

i never heard about " realESRGAN " .. and i will investigate...

BUT what about output  at 1920x1440 ? its' necessary to resize? because i'm little worried about doing some kind of interpolation on original footage...  i leave the output at the 720x576 specifing a reproduction using 16:15 format (to be sure that VLC or other players play it at 768x576 making circles looks good)
upscaling is creating more datas to store .. i was thinking leaving the footage at  original size is better to make "future" players able to upscale using new algorythms (on future... 2040..2050)


posting a frame (on left... the output of OBS studio... on the right the output of Hybrid) looking at it the Hybrid deinterlaced: the "flyer" on the glass is readable.. but was not readable on original VHS played on a TV.

(also i'm not cropping the image to not loose the 4:3 ratio... )