Guys, pls support with vote my idea Topaz AI Trainer - Empower Users to Train Their Own Models
Thanks for the info, I tested this out today with turning off e-cores for ffmpeg and only using them for Video AI exe but didnât see any improvement using only p-cores and setting priority to high and even tried realtime.
The only thing I noticed when exporting is the fps timings didnât stutter as much as wore consistent.
Hopefully the next gen GPUs will speed things up, not long now.
Cropping save not only space and compressing ressource but also allows displayer to adjust perfectly to diffuser using metadatas Mastering Display, coded_width, coded_width, displayed_width, displayed_height
Also leaving 2 to 5 pixels borders help a lot with models that scale up or down the image, so that you can crop them into final displaying, using coded_x
You avoid some borders ringings like this
Is anyone having the issue where they cant use the software at all? Topaz isnât importing any media at all on both Mac and PC.
Can someone explain to me why when I upscale a videofile from 720x576 4:3 to 1920x1080P 16:9 the render output file is 1.942 instead of 16:9??
Settings is:
Codec settings output
QuickTime V210
Profile Uncompressed YUV 10-bit
Container .MOV
Output resolution
1920x1080 FHD
Pixel type Original type
Crop setting Letterbox/Pillarbox
Video type
Progressive
AI Model
Proteus
Framerate 25fps
MediaInfo
Width : 1 920 pixels
Clean aperture width : 1 920 pixels
Height : 1 080 pixels
Clean aperture height : 1 080 pixels
Display aspect ratio : 1.942 ??
Clean aperture display aspe : 1.942 ??
VEAI 5.3.6 has a little regression bug with zooming/panning the video Iâve reported before, v5.3.5 was fine. It might be not important anymore though, as mpv is going to be in charge of the playback soon in the beta, then in release, I hope. ![]()
No because itâs no where as simple as you think.
And itâs not a competition on here to get votes.
Weird. At 1920x1080, Display Aspect Ratio should be 16:9, of course (1.77)
This sounds to be a bad use of DAR and SAR
MediaInfo works with internal tables to ensure an âalwaysâ result
If meta are missing, this can lead to incorrect aspect ratio, but will not lead to unplayable file because displayed_width and displayed_height are searched ahead.
Keep in mind that TVAI use not only tweaked ffmpeg but also return incomplete metadatas.
Hello.
That is one thing we had noticed when we had TVAi.
We didnât spend enough time to troubleshoot but mediainfo showed a TVAi processed HDR project with no HDR metadata but it did show the REC.2020 info.
I recall (somewhere on here) that TVAi supports HDR processes but it doesnât show up in mediainfo correctly. ![]()
Why do you work only with UI for years? No models acceleration at all. It is pretty ridiculous when it runs with 5-8 FPS (nyx, artemis) by upscaling 1080 to 4k with RTX 4080, it takes half of a day to workâŠ
It always did thatâŠ
but at the cost of diluting your purchased content that cost ~20 to 40
Oh, and not using my TV to its full extent?
Of course it has to be done with careâŠ
Be careful not to chop heads and so onâŠ
Sometimes I decide not to crop and sometimes I crop somewhat to at least minimize the black bars.
Itâs not (only) about saving processing time. I also paid a lot for my TV and want it to be used to its full extent.
The 4:3 format wasnât invented for the best viewing experience.
It was âinventedâ to make the CRT as square as possible to minimize the chance of implosion.
Also the way the tube illuminizes the phospor is much more dificult with a wider aspect ration.
If technicians had it their way (so to speak) the thing would be square.
When flat panels emerged they agreed to a better Aspect Ratio that conforms to your perception of the real world.
So in itself 16:9 is a much better viewing experience without question.
We canât however go back in time and tell the directors to reshoot everything to 16:9 or at least tell the cameramen to take care of getting important stuff at bottom and top.
Most of the time everything happens in the centre.
The invention of the 4:3 format had nothing to do with CRTs. The format dates back to the 1890s when it was selected by the Edison and Eastman companies for the production of moving pictures on 35mm film stock, and was the global standard for movies until widescreen was developed in the 1950s.
The first CRTs were round. The first squared-off TVs had round tubes and a 4:3 mask that covered parts of the screen to make them look rectangular. And when squared off cameras and screens went into production, they were 4:3 because that was what people were used to seeing at the movies.
Widescreen CRTs were made for a brief period, but they were very heavy, very expensive and very late, coming out within only a few years of LCDs obsoleting CRT TVs entirely. By that time there was nothing still being filmed in 4:3 format anymore, so except for the very earliest models, LCD TVs have all been widescreen.
Fact remains that the 4:3 was a trade-off of what was technically possible for CRTâs.
It would have been impossible at the time to make widescreens to react on the movie industry starting to push cinemascope to differentiate.
I do admit that I was under the false impression that widescreen was already established in the 50âs. Probably because a lot of old material was later turned into widescreen.
I now know the movie industry started pushing widescreen in the 50âs to differentiate.
It would certainly have been more difficult and expensive to produce a widescreen CRT in the 60s, but the real reason it didnât happen sooner was that there simply wasnât a demand for high end TVs, which was what a widescreen color TV would have been. And when that demand did emerge, it was for much larger screens than could ever be made as CRTs. And most of them were made as 4:3. 4:3 remained the predominant CRT format until the LCD obsoleted the CRT and essentially forced universal widescreen onto consumers.
Not to hijack your point, but I canât even upscale a video like this to 4k. See Topaz Video AI 5.3.5 - 5.3.6 - #46 by meimeiriver
What happens is, that TVAI internally first upscales the image to such a ridiculous size, that blotting starts to occur again. See Proteus 3 black splotche blotting
This issue never seems to get resolved, no matter how many times itâs getting brought up. (And not just by me) Itâs getting rather frustrating, tbh. At least support VapourSynth then, so I can properly pre-crop myself.
EDIT: Haha, Iâm such a moron! I had completely forgotten my own makeshift solution to this issue: VapourSynth Preprocessing with AVFS
Now I can setup a mounted VapourSynth script, like this
import vapoursynth as vs
core = vs.core
core.max_cache_size = 65535
vid = core.dgdecodenv.DGSource (r'c:\jobs\alien.dgi')
vid = core.std.CropRel (clip=vid, left=234, right=234, top=132, bottom=132)
vid.set_output ()
To be upscaled by TVAI, pre-cropped, without first getting blown up beyond 4k.
Often people seem to forget that, the way movies are filmed, on-location camera viewers have demarcation lines drawn on them, for 16:9, wider even, and 4:3. Thatâs just so producers can decide, later, what kind of cut they want. Tl;dr: most movies are shot in a way that already allows for a possible 4:3/16:9 cut; aka, the detrimental effect of cropping off on the sides is often far less dramatic than you might think.
During my high school years in the 1970âs, I was fortunate enough to have 27 inch console TV in my bedroom, which was perched atop my chest of drawers. That damn thing weighed almost 200 pounds!
Nothing about film history is clean or especially logical. Iâm pretty sure Disneyâs Snow White and the Seven Dwarfs was not on 4:3 film and also not what would be categorized as widescreenâand that was made in like 1938.
Between broadcast TV technology and film, video storage has become one of the most convoluted things in existence.