Instead of selecting it from the list it lets you cut the existing text and type it in. Though they should add it in the list of selectable frame rates.
Also in some of the posts above people have mentioned the picture quality not being as good.
With the updates to the app/models couldn’t we be given stats about the quality of the ouput. ie. currently we can test how well the speed has been improved by using the benchmark option but there’s no stats that show how accurate the app/models are in a standard set of tests so we can easily get an idea of how much of an improvement it is (if it is an improvement - or how much quality might have reduced for some things).
eg. couldn’t Topaz create a standard set of test video files (with some for testing upscaling models and some for testing interpolation models) and use them to give us numerical stats for each new app/model release?
They could have a few different types of content, and they could have a real 4K version of a number of different test videos and a 1080p version created from them and then the app could use the different models to upscale the 1080p version to 4K and then it would check the differences (signal to noise ratio/PSNR) of the upscaled to 4K versions against the real 4K versions. And all these test videos would be ones that weren’t part of the training set used to create the models.
Those should be useful stats about the different releases (especially if the PSNR values - or one of them - closely matched the human visual system - so that where we would see a big difference between images/videos that that number would also give a big difference).
ie. they could release something like (or we could maybe run a test option within the app ourselves to create these):
Topaz Video AI v3.2.0, Artemis [version no.] 4x scale, Test Clip(s) 1, PSNR :[numeric value showing quality compared to actual 4K version of the test video] …
Topaz Video AI v3.2.1, Artemis [version no.] 4x scale, Test Clip(s) 1, PSNR : [numeric value…]
My RTX 2070S uses a default setting of max. 83 °C. I reduce it to 75 °C or 70 C°. This lowers the power consumption from 200 W to 135 W by a similar processing performance (for couriosity). I also deactivate the Intel Turbo Boost to lower the CPU Watts power draw.
Veering off-topic here a bit, but how much does your gpu’s draw power while idle?
It should ALWAYS be around 30 watt max!
If it draws something like 100-120W you have a power issue. You can also recognize this easily by your idle Mhz, if it’s over 1000mhz, that’s not idling. My current power draw is 24W with 210Mhz gpu.
I did this by installing this, it doesn’t matter it says multi-monitor, I have only one and same thing:
The changes of color and definition in the version
v.3.2.0 should be thoroughly reviewed.
I attach a graphic test with a deinterlaced TV.
With the help of Vectorscope I have verified various video and photo sources and the problem is constant.
A professional colorist should find solutions or, failing that, as occurs in the main image editing software, add the possibility of typing the ICC by the user and keep it constant during the render.
At what internal point is the color change taking place?
Greetings
The color issue must be resolved.
I recommend that you process a white sheet to be able to deduce the color changes in the final result.
If you have a large Kodak gray chart it will be very useful.
I read your interesting contributions on these errors and some computer prone to color blindness - as you well know has settings in Adobe - reacted inappropriately.
In EFX studios, programmers and supervisors tend to have very fluid lines with the DIT so as not to have post conflicts with the Color team.
Topaz has a great photographer on the team… I understand that with so many programming and development problems he’s going on safari…
Your correct approach is a necessary contribution.
Greetings
To be honest all my GPU’s I’ve overclocked in the past will sit around 75% of it’s clock speed at “idle”. This card’s TDP at “idle” is 144w on the desktop until I remove my Overclock settings back to default. I found It’s just a drawback of doing it. That or it’s my OS being strange… I’m not fussed on power usage personally, it’s the only constant power intensive appliance in my house apart from the fridge and freezer.
Edit: I should mention I have a 240hz monitor and a 144hz secondary monitor. Which may be the increase in power usage at idle.
For my card maxed out at 112%+ power limit using 460w peak TDP will barely touch 80C under every circumstance I throw at it. rarely even gets to 74C on hot days… Whereas others with the same card have complained about it getting to 85C easy so I guess it just depends on the environment it’s in. Trust me 75C is nothing compared to Hardware I’ve had in the past that would be a constant 88C-97C. it also helps to get lucky with a decent binned chip with your series of card.
Sounds to me like your download is being blocked by something. Try temporarily disabling windows defender firewall, and also disable any other antivirus/security/firewall applications you might have.
In December 2019 I was running a TVEAI processing. While the processing was still running I put Windows into sleep because I wanted to sleep too. Next morning the PC was dead and did not wake up anymore. It got replaced the GTX 1070 ti by an rtx 2070S and the power supply both on warranty after only 1,5 years …
Is there a new location for the models folder? Using this guide: Model download tutorial - #29 by dimsimjim, there is nothing else than the logs folder when going one level up. I cannot find any models folder in Roaming, Local or ProgramData. Where are we supposed to unzip these files?
Also, can I downgrade to the previous version? I just want to work for now…
Nothing else but the noise reduction slider was applied.
I believe this was at 50, but it was necessary to strike a balance between 1) enhancing desirable fine detail, 2) preventing the “painted” look of compression reversion, and 3) suppression of random analog noise. If it weren’t for the oversharpening (which is also giving that woman a nasty lip and chin injury instead of a minor shadow) it would be fine! 3.0.4 didn’t do this mess.
Even the blank walls in another part of the picture had “flurries” moving around on them. The “shoeprint” on that white seat back cover is an artifact, yes. But…the dark black edge on the right side of the cover is NOT an artifact…that’s sharpening that should never occur in any noise reduction function.
I did the same test in every version going back to 3.0.4 …where it stopped “sharpening” and actually started reducing the noise. That’s also the most recent version where that “shoeprint” artifact didn’t show up.
P.S. A video has been added to the dropbox as requested by the moderator, however I trimmed it to 30 seconds of the relevant portion–if there was any reencoding, it may possibly alter the results in testing. Quicktime doesn’t tell you if the trim function reencodes or not. If you want the much larger original video please let me know.
Set up your export job, then go to the menu → process → select show export command, then copy it, then go to menu again → process → open command prompt → right click to paste in the export command, press enter to start it.
When the job fails, you will probably see a more meaningful error message. Paste a screenshot of that here.
I’m currently using v3.1.8. Can I install v3.2.0 without corrupting the previous version (install in different folders), so I just test it. I know I can do it with v2.6.4 and v3.xx, can I do it with different version of 3.xx?
Does the command prompt eventually error out after this section? Usually the error messages will appear after this section. You might need to wait a while for it.
The JSON files should be extracted into %PROGRAMDATA%\Topaz Labs LLC\Topaz Video AI\models (Win) or /Applications/Topaz Video AI.app/Contents/Resources/models (Mac).
You may switch back to a previous version by downloading the appropriate release from the Releases forum.
