I asked a manual way to crop via mouse on the beta facebook page, one of the dev replied to it (Matt).
1 Like
As you probably know, searching for a post on Facebook can be tougher than it ought to be.
I would love mouse crop selection, but having implemented it years ago in one of own graphics programs Iâm aware of how daunting it can be. I would be perfectly happy if they let us set top and left along with height and width. It would only take a few seconds longer than using a mouse. Iâm not saying it would only take seconds for them to implement it, but that it would be much easier to so so.
i agree with you.(on the crop part, not facebook 
(i agree on that too lol 
)
I am having an issue with THEIA fidelity padding time to files on upscale.
I do not process audio with the file, and I am procesing to MP4, but the files are adding 2-5 seconds, which when remuxing audio is causing sync issues.
With shorter files of less than 40 minutes this is not an issue, but 42+ minutes and the file is padded.
Normally I would not be bothered, but seeing as it takes 13+ hours to process long files with Theia, this is something that should have been caught well before Theia was released.
My source files are CBR, progressive, and I have tried this in 1.5.1 and 1.5.2
Either the algo is defective, the MP4 compression is bad, or something else is taking place to account for this as the same files can be processed with Gaia or Artemis to less satisfactory results but no padding on the run time.
To be clear youâve tried processing the file with Gaia or Artemis in the same software version (1.5.2) with every setting the same except the model and the length of the video is consistently different?
I did that in 1.5.1
tested Theia again in 1.5.2 and got the same results, so didnât bother testing Gaia and Artemis.
I find the 42+ time gets grey, as in intermittent results showing padded time, when the time is in the 42-70 minute range. Where files that are ~80+ have consistently have their run time padded with Theia processing.
I recently did several files in a batch using Theia, but none were longer than 55-60 minutes and there were no issues with the run time. But when I did a single file that was ~77 minutes long, the run time was padded by 3 seconds at the end of processing, And a 102 min file was padded by 5 seconds.
It is rather time consuming to rerun files of this length numerous times.
I really like the output results of Theia, but this time padding for long video files is a deal breaker, so far.
PC Specs:
Ryzen 9 3900x
32GB 3200CL16
X570 Aorus Master mobo
EVGA RTX 2080 TI FTW3
OS: 512GB Samsung 970 evo
Storage: 2x 2TB Seagate Ironwolf 7200 RPM HDD
Custom loop cooling on both CPU and GPU. During processing the GPU rarely gets hotter than 54C.
It is likely occurring on all models then. Switch the output type to an image sequence and merge them after. As long as the input video isnât variable frame rate (which we donât support at all) this will work reliably.
@onemob Yes, VEAI does not have any tolerance for variable framerates whatsoever. You will get strange results every time.
Are there any functional differences, apart from file size, between the 8 and 16 bit tiff?
Does the AI natively output at 16bit, then 8 bit is the down sampled version etc?
I should have noted CFR, not CBR.
For example:
Width : 716 pixels
Height : 480 pixels
Display aspect ratio : 16:9
Original display aspect ratio : 16:9
Frame rate mode : Constant
Frame rate : 23.976 (24000/1001) FPS
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Progressive
Or:
Width : 720 pixels
Height : 480 pixels
Display aspect ratio : 4:3
Frame rate mode : Constant
Frame rate : 23.976 (24000/1001) FPS
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Progressive
The models actually output signed 32bit float right now and that is converted to 8 or 16 bits. If you were happy with the results from MP4, JPEG is all youâd need. It looks better than MP4. TIFF would give a step up I imagine though it depends on your final use if that will be noticeable
I tend to use PNG output, JPG can introduce artifacts of itâs own.
(Granted I did buy an extra hard drive just for VEAI lol )
With Nvidia announcing new cards today, with some more that twice the speed of RTX 2080 Ti, like the RTX 3090, will Topaz due more to utilize the processing power of these cards, to reduce the render time? What wears on me more than anything, is the 5 hours it takes to render 30 minute standard definition to 4K, and thatâs utilizing RTX 2080 Ti. As time goes by, Iâm becoming less and less interested in spending this amount of time to upscale a video, let alone an entire television series. To be honest, the excitement is gone, and it seems more like work. If I could take that 5 hours and condense it down to an hour, that would huge step forward!
1 Like
OK, a 102 min file, CFR, rendered out to image set as PNG using Theia Fidelity did not pad time.
However, the same file when rendered using Theia and MP4 did pad time.
Is this acknowledged as a bug?
It took 16 hours to generate the results on my end, have you tested it as well to see what sort of results you get?
Yes MP4 isnât great as an output container. It has never worked quite right. The library we use just doesnât quite read and write files right all the time. The good news is we are working on overhauling our video IO pipeline. The bad news is it takes time and is a big project. For now we highly recommend only using MP4 for previewing video and using one of the image sequences formats for any serious work.
1 Like
My concern, and this is with the caveat that I am extremely interested in the 3090 to speed up my processing times, is that the 3090 appears to be classed as a Titan; based on the 3080 getting billed as the flagship card.
Enter speculative thinking: If the 3090 is classed the same as a Titan, then VEAI might not support it as Titans are not supported now.
Until the cards are released and the devs actually buy them, I doubt we will have any answers.
Thanks. The new video IO pipeline improvements are something I can patiently wait for, knowing they are coming.
Rendering to image set does take a bit more time, but not having to redo projects takes less time 
I was in the process of planning a new computer for VEAI, and the Titan was on my shopping list. I canât believe theyâre not supported. Why?!?!
Edit - It should work with a Titan RTX, right?
TITAN cards not supported ? How could this be even possible ?? They are just the same cards with full specifications. VEAI uses CUDA for GPU acceleration so as long as any card has CUDA cores then it should properly work and the more CUDA cores then the more speed processing it should have.
How can I turn off face recognition in Video Enhance AI Gaia CG ?
The Program is distorting faces in low resolution x2 upscale processing.
I can fully recover facial features in high quality with external de-noise.
Any workaround to disable the face recognition features?
1 Like
Thats actually not true. Im using a Titan and it works, however I am still not satisfied with the performance on VEAI. But that might be because of other factors that I still not know
1 Like
I had an RTX Titan last year and traded down to a 2080Ti, as I had later found out I couldnât get the full performance out of the Titan due to throttling. It was only topping out at 80% GPU usage and 50% VRAM usage as it was hitting 85 degree celcius while doing upscaling. So I would (if your case & motherboard allow it) ensure the card has plenty space either slot of it, and that your case fans are not set to âquietâ mode, and are instead set to at least balanced mode, while considering putting them up to âextremeâ for the time periods youâll either be encoding, or running another GPU heavy task to see if that makes any difference for you.
No doubt itâs a great card, but my experience tells me most home setups canât provide the cooling needed for it to use them to their max. Iâm hoping these new 30 series cards with their new cooling solution will address such issues, as I definitely want to get at least a 3080. Might stretch to a 3090, but thatâs going to depend on how well the cooling does, and if I can get a decent sale price for my 2080Ti, which sadly doesnât look promising after the price points Nvidia have put out!
