Ai Enhancing and Upscaling

[magnu]

Apologies if there may be a similar question to this elsewhere, but it seems here is the best place to ask.

I have been using the Topaz Video Enhance Ai software to upscale and improve the quality of some of my videos (though on some occasions I use other software such as Vegas Pro and NeatVideo to clean up the quality), and on the whole I have had good results.
I also have DVDFab with all the bells and whistles and it also includes similar Ai software yet, even though I own the software, I haven't had the urge to try out the Ai side of it since I've never came across anything on the net talking about it?

So it got me thinking. Has anyone here tried the DVDFab Ai software or seen any reviews on it anywhere, hopefully with comparisons to Topaz? Also, has anyone tried a mixture of both programs and achieved any good results, compared to just using Topaz?

[lollo]

Sorry if my question is a bit off topic, but I experienced a problem with Topaz Video Enhance AI about the proportions of the video when upscaling.

I do it rarely, but in my AviSynth flow I upscale from 720x576 4:3 to fullHD using:

nnedi3_rpow2(rfactor=2, nns=4, qual=2, cshift="Spline36Resize", fwidth=1440, fheight=1080)

and add 2 borders of 240 pixels at left and right.

On Topaz AI I simply choose HD (1980x1080) option in the menu.

When I compare the final results the image is quite different in width (checked with VirtualDub and VLC).
I also experimented some different cropping to the source video (i.e. to only work with the active pixels, 704x576, 702x576 and even lower) but the difference is still there.
Is there a setting in Topaz AI I am missing or something in my flow is wrong?

original SD frame (vdub, DAR 4:3 not respected!)

[Attachment 58857 - Click to enlarge]

vdub upscale comparison, Topaz on the bottom

[Attachment 58854 - Click to enlarge]

vlc upscale nnedi3_rpow2 + borders

[Attachment 58855 - Click to enlarge]

vlc upscale Topaz Video Enhance AI

[Attachment 58856 - Click to enlarge]

[davexnet]

Topaz is probably not taking the DAR of the source file into account, while you are so doing in the AVS script

[lollo]

I think you are right: Topaz probably tries to read the DAR in the input file. But the input file is an avi generated from HuffYUV capture + AviSynth + Vdub, so the input DAR info is simply missing, and Topaz assumes something else than 4:3 or has a strange 16:9 or whatever else default.
My mistake. I will try to add the DAR info to help Topaz...

[lollo]

Update:
from Topaz developer: "The AI processing can’t handle non-square pixels so we have to use the aspect ratio in the metadata to convert to a different resolution with square pixels." Together with the missing DAR info in the input file, this can explain the problem.

Solved forcing DAR on input file; I used ffmpeg -i <file>.avi -aspect 4:3 -c copy <file>_dar.avi


[Attachment 58870 - Click to enlarge]