DVD is definitely 8 bit only. Even 1080p blu-rays are only 8 bit.
But, if VEAI converts to 16 bit internally during its upscale/enhance process, and uses the extended color depth during its processing, and then keeps this additional data throughout the entire pipeline, it might make a difference if you are targeting HEVC 10 or 12 bit, or doing other processing afterwards that supports the additional color depth. The question is whether VEAI is actually doing this, or just writing 8 bit data in a 16 bit format.
Also unclear is why they donāt offer a 16 bit compressed format. TIF and PNG compression is lossless. I can think of no rational reason not to use it unless you have a very slow computer. Which presumably you donāt if youāre doing any of this stuff in the first place.
Even though the source material is 8 bit, since thereās extensive processing happening, the additional color depth can still be used by that process and make better decisions.
Itās really easy to demonstrate this by taking a high quality 8-bit source (e.g. a blu-ray rip) and encoding it to HEVC 8 and 10 using the same settings. The 10-bit encodes will show noticeably less crush/banding in dark scenes. Itās not intuitive why this would matter when the source is 8-bit. But when youāre doing any kind of processing, completely new data is generated from the input. If the processing algorithm has a larger color depth in which to produce its output, it can make better choices.
viktorz3008
(Z...z...z š“ I'm sleeping. And I love Sumire š.)
42
I think TIF and TIFF are more common than other outputs. If the input is 8 bit, then thereās no reason to choose 16 bit TIFF. I believe VEAI will create color bands all over the footage because color format doesnāt match. I donāt think it will help. Or at least does not make any difference. TIFF is just a waste of time and resources for movies.
Again there is a reason, if VEAI uses a 16 bit pipeline internally. Once you start manipulating stuff, the extra space matters; and if your final target is HEVC 10 or 12, youāll keep that extra detail. This is why they master movies in 4K and 8K even though theaters are only 2K. Each time you process a frame, you are limited to the resolution and color depth of your medium; having more during that processing will result in less degradation and better choices by an algorithm at each step.
But none of that will matter unless VEAI uses a 16 bit pipeline internally which I donāt think we know.
viktorz3008
(Z...z...z š“ I'm sleeping. And I love Sumire š.)
44
The problem is, the raw movie which is 2K with RGB 14 bit or higher I believe. They compressed it down to 10 bit. Itās possible to downscale, but upscale 8 bit to 10 bit? I donāt think so. It could be worse than original 8 bit.
Suppose I am doing math using integers as input, and integers as output, Iām going to start with the number 2, then divide by 3, then multiply by 3.
With only integers during the pipeline:
2/3 = 1
1 * 3 = 3
Final output ==> 3
With one decimal point during the pipeline:
2/3 = .7
.7 * 3 = 2.1
final output => 2
Wouldnāt you say that the result of using one decimal point during the pipeline is a lot better than not doing that, even though both the inputs and output are integers?
Processing images is just math. This is the kind of error that can happen. So if VEAI is using 16 bits internally, I definitely want the extra resolution it used during processing.
Working exclusively with TIFFs (got an SSD now) in 16 bit uncompressed actually helped to prevent banding.
I also added a āsmallā amount of grain with each Topaz VEAI step and - done! I no longer need to add a deband filter with DaVinci Resolve afterwards.
This is actually very important if you should have a video/series/movie which may have one of the
following characteristics:
Areas of largely the same color (or shades of the same color)
Close-up shots of faces with changing light levels (like someone talking in a dark room/scene, where one area of the face is visible due to light and the other less visible for being in shadows - the transition between light and dark can sometimes be visible through āhard linesā when banding appears)
Quickly changing lights (like explosions, flickering lamps and such things)
Frequent changes between live action and cgi/vfx elements
Complex or fast movements (when banding occurs here, it can give a sort of streaking look on lines)
At least thatās what I have observed with my encodes.
Cool! Very glad to have confirmation on this, I am going to switch. Iām 12 bits throughout my pipeline before VEAI, and encode to 10 bits, so seems like I should benefit too.
Whatās the best method here for something like a music concert in dvd with different lighting etc and in 4:3.
Iāve got a David bowie concert from 2000 and I ripped with dvd decripter and separated the video and audio with ffmpeg.
I than loaded the video in VEAI and upscaled 200% with gala hq.
I than imported into filmora and did a scene detection and adjusted the color etc for each scene than exported that out for the enhanced 4:3 version.
Went back into filmora and adjusted the crop for each scene to 16:9 than exported that out again but in 16:9 widescreen.
I than imported both into VEAI and cleaned up the videos with Artemis strong dehalo to at 100%.
Youāre going to be disappointed no matter what you do. You should go into this with reasonable expectations.
These processes and tools can result in significant improvements in perceived quality versus watching the original source, but itās never going to be blu-ray quality. The source material just doesnāt exist.
Sorry I should have rephrased that not obviously bluray quality but significantly enhanced I should say, there are some long distance shots where the camera zooms in from a far and the faces are very pixilated so I know Iām not going to resurect those faces as no detail to begin with.
Sorry to have to tell you this, but user ājamietreā is correct: You will NEVER achieve a level of quality that is comparable to BluRay quality, at least not working from a standard source.
In general, anything that is being fed into Topaz which is not at least 720p resolution and encoded at a āreasonableā level of picture quality, will certainly never come close to anything BluRay-like.
I have experimented with feeding higher quality sources into Topaz and that can result in amazing quality that ācouldā rival a BluRay, but those sources - generally speaking - were mostly already in 1080p, some were in 720p but at good quality (WEB-HD, not HDTV).
When working from DVD / SD-resolution, the best you can achieve is what I would call a āDVD-restoration/cleanup levelā, which would basically be a reasonable compromise between āsaving what is there alreadyā vs ābuilding up/enhancing what could have been thereā.
So, upscaling with Topaz (and possibly other software) and then playing the upscaled material on a device that also does image enhancement again ācouldā result in a well-balanced visual quality (clean, sharper, more vibrant color) - but you will also have to live with the possibility, that problems and errors present in the source will at some point become visible since they āmightā be enhanced as well (like strange looks on objects or people in the background or elements with low resolution in general).
My upscaling project on Trek took about 14 months to get my workflow to the point to where it is now ā¦ I have started over many times. But if you only have a few videos/files, then I guess you should be able to do it with less time involved.
I am not saying this to put you off, I consider my efforts worth it every step of the way, since many improvements are possible, but every workflow is individual, since every source is different.
Thanks guys for the feedback. Was looking around the net for AI facial reconstruction from a blocky pixals to a high resolution photo, unfortunately only for photos at the moment but itās quite exciting if it could be implemented in video.
Itās trivial to export a video a series of images, so thereās no reason you couldnāt use this to process a video if that was a goal. But, looking at the project, it doesnāt seem like it would be suitable for just passing in any image that happens to contain a face somewhere (e.g. as part of a long video), I think it works with images that are just pictures of a face.
Interesting for sure, but I also think that it is currently usable for faces only.
Furthermore, the requirements say that you need CUDA-drivers and a compatible GPU, so thatās a limitation, since not everyone will have the hardware.
Hey there guys, I thought I should give you proof that working from uncompressed TIFFs can help combat banding and result in less visible issues due to 8-bit in your workflow/pipeline, which eliminates the need for further debanding filters down the road (which also helps to retain more detail).
The samples are from the pilot episode of Star Trek DS9 - I deleted the old ones I posted here before for reasons of storage capacity in my dropbox.
So, I will provide you with new comparison images between the original SD-resolution DVD image (on the left side) and my 4K upscale (on the right side). Please remember, my player strechtes 4:3 to fit the screen and I did cropping left and right on the upscale, so the right image will seem to have more width, but the resolution I rendered it out in is 2880x2160p (which is 4:3 in 4K), so one could easily adjust to display in 4:3, which would restore the original aspect reatio - I just wanted to have the option to watch it without black bars left and right.
So, here is the link to the images (careful, about 250 MB and I think about 215 images ; password is DS9Up):
Also, since the work canāt be truthfully judged just by looking at still images, I decided to include a longer video sample in 4K (the first 16 minutes of the pilot, which has scenes covering everything from close-ups over distance shots, live action, vfx/cgi, slow and fast moving scenes as well as light changes and bright vs dark scenes). Here it is, same password as before (file size: about 2,5 GB):
Also, I did some adjustments on my workflow, according to the new procedure involving TIFFs, you can have a look at it here (pdf-file, same password as above, this time better quality for reading):
Hi there! I appreciate you sharing your work with us. Very nicely done! Iāve also done a little bit of work similar to yours, and I think my input might be helpful for the group. In short, I also do some work upscaling old DVDs to 4K using VEAI, thereās a movie in particular I am working on now that the Blu-ray 1080p American verison is hard to come by, itās a 1985 movie - Gotcha! First I purchased a new computer - the Lenovo Legion T730 Intel i9-9900K 32GB RAM NVIDIA 2080 512GB SSD 1TB HDD the reason I purchased this computer was because it has a DVD/RW. I didnāt care for that optical drive, because I removed immediately once the computer arrived and swapped it out for a 9mm 4K Blu-ray/RW. This PC now is a full fledge 4K beast, meaning I can burn and edit 4K movies.
Okay back to VEAI: the steps I took to remaster the movie. I needed to rip the Gotcha DVD disc and used MakeMKV but this resulted in too many sec./frame when loading into VEAI. Solution I used DivX Converter and upscaled the video to 4K. DivX Converter can also rescale 4:3 video to 16:9 or anamorphic scale easily and without cutting out any part of the image, a excellent feature that would be wonderful if VEAI could incorporate?
Once DivX Converter completed the conversion, I uploaded the file to VEAI and the file sec./frame were reduced, but it still took my computer 6 hours to complete the entire movie but it was done and in 4K.
I just wondered if anyone has tried applying hdr to their upscaled dvd rips
I downloaded devinci resolve 17 (not the studio version as you need a serial number) and allthru quite complicated to get my head around I was watching some Yt tutorials on Devinci resolve 17 and demonstrating hdr being applied with a oled TV obviously using a hdmi output from the laptop.
Loads more brightness and color on the oled that you canāt see on the laptop screen.
I guess you would have to apply it differently to every scene.
I am not sure if applying HDR to a āstandardā source would be practical or even visually advantageous - I have seen many examples around the web from professional groups as well as fans of the material in question.
I suppose it would have to be done scene by scene with painstaking precision, otherwise it can be hit and miss.
Color grading is a similar matter.
Personally, I do not have the experience to talk about HDR, I do color grading in the sources original range.
Hey there community, I wanted to share another little example of my work.
Since it is well known that the later seasons of Star Trek DS9 (and Voyager as well) had better quality when it comes to the production values of those shows and conclusively their DVDs, I wanted to see if my workflow would hold up against the later seasons, so I decided to do two test scenes from the DS9 season 6 episode āSacrifice of Angelsā.
Both scenes focus more on the cgi/vfx elements, but I included a few shots of live action/actors/faces as well.
Please be advised that both scenes had a lot of camera shaking (which is logical, since theyāre both battle scenes), so it was hard to find images that were actually āstillā.
