it’s a good job! to go from a low definition to such a high definition, inevitably that creates a blur but there is still all the details, and for me a correct level of noise. besides, there are no artefacts, it must be nice to see compared to the original! it really looks good!
Thanks. Yeah, the resolution … that’s always something in the back of my mind. I mean seriously, no matter how bad or decent the original looks, it’s still basically scaling up 400%, so every little mistake will show.
Also, interesting fact about working with Topaz upscalings: There is a reason why I first re-scale my material within StaxRip - the original dimension does not conform to a “round” scaling factor (if I were using the original with automatic crop, I would end up with uneven scalings like 180,xx % instead of 200%). Once I adjust, it goes 200% towards 1080p in the first upscaling run and again 200% towards 4K. That’s important, otherwise Topaz VEAI apparently does two scalings.
Example: If I use a strangely cropped resolution and upscale with a factor of let’s say 180,xx%, then Topaz will first do a 200% upscale and then use bicubic resize downwards to 180,xx%. That can (can - it does not have to, but I have seen such evidence and there are discussions in various topics around the forums) introduce additional artefacts.
And when you think that through, if I ran the source material twice through the upscaling, that would actually mean 4 times (including the downward steps). I guess that could be a reason for artefacting that conclusively would show even more at higher resolutions.
Yes indeed! I just paid attention to this detail! my videos are 720x576, until now I was rendering them in hd, (x125) so I just rendered them to x200 (1440x1152) to see the change.
Hey guys, sorry but I had to delete my samples since linking them around here is not allowed (copyright and so on). But I will create another topic in the showoff spot from VEAI, just search “Upscaling old DVDs to 4K (SciFi & Star Trek)” and we can talk in more detail there, since my attempts have really been coming along quite nicely.
In case anyone is interested, just search for the above mentioned topic.
Sounds quite interesting. I am working with PAL as well, so I feel your pain - then again, lately I have discovered that Babylon 5 DVDs are actually worse than Trek-DVDs in terms of how the VFR between live action and cgi/vfx was handled (yes, it actually is possible to somehow do worse than something that is already bad).
At the moment, I have discovered a sort of “multi-software approach, coupled with a multi-model approach by Topaz VEAI”.
In “short”:
- Extract the original raw data from the DVDs
- Extract the original audio tracks from the DVDs
- Adjust / slow down the PAL audio from 25 fps to 23,976 fps (kills PAL speed-up)
- Use extensive filtering to get rid of issues like: Rainbowing, Interlacing/Combing, Block Artefacts, Halos, Ringing, softness (add sharpening), Banding - also: slow down video to 23,976 fps
- Use Topaz VEAI to enhance and upscale
5.1) Step 1 - Gaia HQ v5 at 100% Denoise/Deblock
5.2) Step 2 - Artemis LQ v12 at 1080p (custom to save 4:3 ratio) - careful: you have to actually make some sort of “scene-check” afterwards (in some cases I replace shots with Artemis MQ because it handles elements like nature and fire better)
5.3) Step 3 - Gaia CG v5 at 4K (custom to save 4:3 ratio) - Use DaVinci Resolve to combine TIFF images back into video and add “finishing touches”
(those include: Color Grading, Neat Video Plugin, slight blur, Texture Recovery and Film Grain)
Wanna see results/samples? Here are some examples:
Personally, I think it’s looking nice, considering what the source is offering … best open the images in a different window at full size.
2 Likes
It must be so painful for the creators of B5 to see their show butchered so much. We’re talking about a show that was produced with the most forward vision ever before - shot on 35mm, all shots framed for both 4:3 and 16:9, all VFX projects backed up and saved so that 16:9 versions could later be rendered. And yet, renders of 16:9 VFX were never ordered, the production company that saved the projects eventually went bankrupt in 2002 and the files assumed lost, and the best rendition we currently have is a 4:3 HD remaster with all effects shots upscaled and converted to 24p with terrible jerking motion.
So true … and sad …
didn’t work for me, as Dione Robust still left too many de-interlacing artefacts in the footage and both models in concert produce severe oversharpening.
This paragraph suggests you may not have understood the rest of that post, so i will go over something I have discussed previously.
I never use Dione or any deinterlacing model in Topaz as an actual deinterlacer. All footage fed into it is progressive. You alter the settings in the model so it treats it as progressive. The source file being fed into it has no interlacing left to deinterlace so if you are getting interlacing artifacts, I am guessing you didn’t use the script immediately before it that removes the interlacing in he first place.
In other words, don’t attempt to deinterlace in Topaz.
Having sad that, I will mention as I see a lot of people not realising this, but no Star Trek Voyager that I have yet come across is actually interlaced source material, whether you are using NTSC or PAL. PAL is the sources I had as well and had to special order the NTSC, specifically because every time I compared them, the image result was cleaner with the NTSC.
Additionally, the vast majority of NTSC episodes contain little to no mixed frame footage making them dead easy to IVTC them. I did the entire season 4 of NTSC and they are 99%+ straight telecined. Episode 1.01 from Season 1 is a notorious example of mixed frame footage but it is the outlier rather than the norm.
With PAL, what they did was convert it to 25fps and then phase shift interlace for broadcast. This is because to broadcast the image, it had to be interlaced or the TV’s wouldn’t know how to display it. The one advantage for us is that it means you do not have to deinterlace the footage using traditional means (like interpolating missing data between frames) as there is no spatial interlacing anywhere.
The easiest way to see this is to simply load it into Avisynth and run a straight TFM() on it and then scroll through the frames. True interlaced footage will retain the interlacing artifacts - Voyager, however, completely removes them. You get left with the original 25 fps before interlacing with no need to actually deinterlace and try and interpret (make up) missing data.
It also means that doubling the frame rate on PAL is somewhat pointless. You basically take the originally 25fps, phase shift them to 50 fields, then try to make up missing data to turn it into actual 50 frames, then convert it back to 25, where the original 25fps with no interlacing at all is already there. It is a lot of work, with a lot of processing for nothing as a 10 second script will restore the progressive footage. Even the D2V generation in DGIndex shows the original is actually 25 fps progressive.
This is Frame 3544 from S04E02 PAL native, with the interlacing still present:
(NOTE: 200% size display for clarity, but I am not resizing them to square pixels so they will both show “distorted” images with differences between PAl and NTSC which are differently sized)
This is the same frame running just TFM() only and no other filters:
The problem with PAL, however, is that to get it to the 25fps in the first place, they had to run it through a black box that converts the footage which has “done things” to the underlying footage that are irreversible. You have some places where interpolating has been performed and the original frames show the ghosted inbetween frames that are missing in the NTSC versions. You have also had some extra processing done to the video so you have lost some additional details as well.
The biggest issue is that because the original telecined footage was conerted to 25fps progressive, it did not entirely dinterlace the source. So what you are left with is an underlying progressive source with interlacing artifacts that are baked into the source frames themselves making them a pain in the ass to remove.
A good example is here:
Those are not from a bad deinterlace from the source, those artifacts are part of the source itself. The only way to remove or lessen them from the PAL source is to then run it through filters that try to recover the details and remove the remaining artifacts.
Santiag() is a good option for that, ie the same frame with that filter applied:
BUT, if I take the same frame from the NTSC source with no filters applied, you get a cleaner starting image without needing processing:
This usually means you can tease out better fine details in the upscale.
I should also re-mention that my goal was to make a better looking 1080p version that tries to minimise artifacts whilst retaining the best small details. The idea, or goal, is not to make some 4K masterpiece that is nigh impossible with the source, but just have a copy that looks better and closer to HD on my main TV. For that, small detail is paramount.
Even since this post I last did in this thread with Dione, I probably have done about 100x new tests - new Topaz versions, new Models, re-testing old Models etc. This did, however lead me down a rabbit hole of paradox. I found that I could tease out insane amounts of detail at times, but it looked revolting moving. I actually had to tone it back and keep checking what it looks like being played back. I could make some fantastic looking still images - that would look atrocious when trying to watch it.
I had to basically accept that I could not look at the images, enlarged on my big monitor, and fix all the small issues - it all came down to how it played on the TV. It also means that my copies may not at all suit anyone else - others may hate it, but it suits me.
https://drive.google.com/file/d/1vnFcMKskSOgHu9FceaEjmRd7Cjqwn_f4/view?usp=sharing
This is an extract from the same scene with the doctor in it. As you can see if you display it on a 2k or 4k monitor, sitting a foot away, you will see all the flaws - it is not perfect and that wasn’t what my goal was. But if I play this on the main TV in the lounge (52 inch full HD), sit mack on the couch and watch, it is 10x times better than the streaming versions, the up close sections are much clearer and many of the distrating artifacts are gone. Also, which was paramount for me, I have a great deal of trouble seeing any “upscaled” artifacts or where it is obvious processing has been done.
IE I don’t want friends to watch and think “that looks kinda weird”. Hopefully, they may watch and not even realise I did anything to the image, and then goal achieved.
These samples look very impressive! I wish I could manage the steep learning curve and take on Avisynth, but scripts and CLI’s have always remained a mystery to me…
Right now I can’t post any of my own samples (5 posts or more required).
I still need a bullet-proof deinterlacing method. Dione is a very welcome addition, but like most (if not all) AI models, it is quite terrible with text, Star Trek displays and so on.
There is a scene at the beginning of “The Way of the Warrior” (DS9-4-01) when Odo changes form and the background shows a huge display “Promenade Directory”.
VEAI can’t seem to get it right.
I don’t have DS9 DVD’s. Everyone is always using DS9 as examples and I can never cross check as my intention was to finish Voyager then look at the DVD’s for DS9 - I am one of those weird ones that always preferred Voyager to DS9.
However, to your post, a quick guide. Make sure that you make a decision to either use x64 or x32 and stick to it because they all interconnect and all require the same version types to work. You also have to make sure filters you install use the right version as well. x32 versions are often easier to find, but I have used the x64 version the whole time and had few issues. To save some time I am uploading my filters which took me a large amount of time to compile, so you may want to go x64, but if not, keep in mind you will need to fine and install filters yourself.
- Download and install Avisynth+ AviSynth+ - Getting Started
- Download and install AVSPmod Releases · gispos/AvsPmod · GitHub
- Download and install Virtualdub http://virtualdub.sourceforge.net/
- Inside the folder where Avisynth+ is installed, there is two folders - plugins+ and plugins64+. The first is where 32x bit filters go, the second is the folder I have uploaded of mine with filters for x64 here plugins64+.rar - Google Drive
This should get the applications sorted, but it has been a long time since I did this, and I can usually work out issues easily as I install as I have been working with programs for a long time, so if something doesn’t seem to work right above, or there is a step I have forgotten, let me know and I should be able to assist.
AVSPmod allows for a GUI that you can insert scripts into and see live previews of the results in the bottom window. It will also give you immediate errors if you did something wrong. If you use the 32x bit version, you may need to find filters from Avisytnth repository and want is the non-64 bit versions of the dll files. The link to filters is here: External filters - Avisynth wiki
If the filter is really old, it may not have a x64 bit version and only be x32. I usually just avoid those. You will know if the download for the script contains no x64 bit folder option with a dll inside.
To load a file in, I will step by step as well, but you may know some of this (potentially) already:
- Download and install DVD Decrypter http://www.dvddecrypter.org.uk/
NOTE: The actual method of getting the files off the DVD can technically be done by several methods. I use this as the resulting VOB is immediately compatible with the next step and makes things easier and so far I can extract any region disk I throw at it - it won’t work with newer DVD’s but for Star Trek from the 90’s it is usually fine. - Download and install DGIndex DGMPGDec 2.0.0.7 Free Download - VideoHelp (its part of this package)
- Open DGindex and load the VOB file extracted. Make sure that in Video > Field Operation settings it is set to Honor Pulldown Flags
- Save project by selecting File > Save Project and Demux Video
The above will give you a .d2v file, the video component and an audio extract which is a native extract (don’t have to extract it again and can be re-used anywhere audio needed).
Then go back to AVSpmod, and assuming you have everything correct, use the following script only:
D2VSource("Path to File")ConvertToYV12()
Where the “Path to File” is the exact path of the .d2v file that you just created, in quotation marks. Mine, for example, looks like this:
D2VSource("D:\Video\NTSC Season 4\Disk 1\S04E02 Demuxed\VTS_03_1.d2v")ConvertToYV12()
If everything was done correctly, you should be able to click the line down the bottom, or press the Toggle Preview button (down arrow) and see a live preview of the video. You could even press play and it should start playing back the video - without sound at this point.
If you have gotten to this point and see a video, used my filters, then 90% of the hard work in settings up scripts is done. If you did use my folder, you will also have QTGMC and a host of other filters everyone keeps talking about here already enabled and in theory, should be able to copy any of these Avisynth scripts and test them.
Now the only downside is this is how to enable and setup scripts to start using filters - you still need to run them through TVEAI and upscale and then combine them and add in sound etc. So its not the whole process, but a fair way along. Extremetech just did an article with an entire step by step guide on re-assembling footage (as well as their method for processing but how much you use is up to you).
So at this point, if you have a PAL source, use the following command only in the script editor and refresh the live preview:
SeparateFields()
What you will get is the individual fields of the image (which will display squashed as they are half fields). I need you to find a spot with sideways motion (where the most interlacing appears) and scroll with a mouse and note the following:
- Does the image move between each frame or
- Does the image stay static but with a very slight vertical shift on each second frame.
True interlaced footage has temporal motion between the fields - that is, that when you scroll one field at a time during movement, lets say a ship moving left to right, you would see the ship move more to the right on every field. If it does, the source has genuine interlacing and will need to be deinterlaced. It will look like this:
Frame 1 - Starting image
Frame 2 - Move to the right a bit
Frame 3 - Move to the right a bit
Frame 4 - Move to the right a bit
Frame 5 - Move to the right a bit
However, every episode of Voyager I have tested has so far had phase shifted interlacing. This will display as number 2 above - when you move through the frames the ship will do something similar to:
Frame 1 - Starting image
Frame 2 - Move to the right a bit
Frame 3 - Move up a little bit (no horizontal change)
Frame 4 - Move to the right a bit
Frame 5 - Move up a little bit (no horizontal change)
As I don’t have the DS9 discs, you would need to check yourself first. Or upload a small section of VOB and I can check.
If the result was option 1, the best Deinterlacer is usually QTGMC and there is a lot of posts about this usage - its a veritable rabbit hole of options. Extremetech have examples, I have posted examples, Codegrey I am pretty sure has examples - take your pick. Note that depending on what you try to use, it may slow the processing a lot - some of the QTGMC examples I have used can reduce processing time down from hundreds of frames per second to under 10x frames per second.
If the result is option 2, do not attempt to deinterlace it. This kind of interlacing is a cheat method for broadcast and its what the Voyager PAL discs so far use. Essentially they take a single progressive frame, and cut it along the interlacing lines and divide that frame into two fields. The result is two field images side by side that are virtually identical but because one field is half the lines and the second is the other half, the image will appear to vertically move ever so slightly, before actually changing on the next field.
What this means is you can reassmble the progressive frames exactly by going back to your script, removing the SeparateFields() line and instead using this only:
TFM()
Your preview line will now show a fully “deinterlaced” version of the source that you can then do other things with.
NOTE: As I mentioned above, if you ARE using PAL discs, it is likely it will have baked in interlacing artifacts from the telecine conversion from NTSC. These won’t be removed and will make the upscale terrible. I would highly recommend, after the TFM() line putting in the line:
Santiag()
Which will usually clean most of those up.
Next some god news. If you are using the newer version of TVEAI - anything from about 1.9 inch and more recent (I think, cant remember when support was added) but the Avisynth script file is directly importable into Topaz. Simply save the script file, locate it in folders and drag and drop the file onto Topaz and it will load the output of your script into Topaz, deinterlacing included. That way, you now have a “deinterlaced” source you can play around with in Topaz.
Now some last notes:
- Large scripts can slow down the PC and can also occasionally crash. As such I always work with images until upscale is fully complete. It also means that if you are doing some heavy processing, it may work better for you to extract your processed file as an image sequence and then import that image sequence into Topaz instead. I typically do this to avoid issues.
If you do, this is where VirtualDub comes into play. Load the saved script file into VirtualDub and then choose File > Export > Image sequence. This will process and extract the video as an image sequence ready for import into Topaz.
- Depending on your PC, if you add this line at the top of your script:
SetFilterMTMode("DEFAULT_MT_MODE", 2)
And then this at the very bottom:
Prefetch()
This will multithread the processing from Avisynth and usually speed things up. Note that Prefetch() on its own will auto choose the number of threads, but you can force it to use specific number of threads by putting a number in the brackets, like Prefetch(4) which will use 4x threads. This can allow you to reduce the load if you want to do other things at the same time.
Good news is if the extract crashes - and it will from time to time - you can set the start and end image points and restart it and it will go from where it left off. Additionally, if you upscale to an image sequence in Topaz, you can do the same thing.
- Now I personally use Premiere to re-assemble the images as its easy and straightforward. If you don’t have it, Extremetech’s guide has a step by step on doing it using I believe free application. There is also ffmpeg CLI but it always takes me a while to get that right and I am not so confident in a step by step for that, so I won’t write one here.
If you treat the extracted images as 25fps and then add he audio as is, it should all sync and work. At this point would be where you would have to deal with changing framerate and adjusting audio pitch etc to make it more normal, but as I don’t do this, I am probably not the best person to guide through that step if you are not sure.
- IF, on the other hand, you have NTSC sources, we have a whole different issue with mixed framerates. Thankfully 95% of the episodes you can ignore this and just use this script instead of the one above:
tfm(d2v=“Path to .d2v File”)
tdecimate(mode=0, hybrid=1)
And you will be left with a base 23.976 source file and everything proceeds exactly the same as above, but treat the footage images as 23.976 instead of 25 and the audio should need no adjustments.
HOWEVER and this is important, some episodes have genuine mixed frame footage and this script will have issues, but there is no way to know if the episode does or not until you run some additional processes.
I won’t cover those here now, but if you need to know this as well, let me know.
1 Like
I will show more samples once I get around to posting them - that being said, Trek DVDs (and SciFi shows of the 90’s in general) have always been and will always be terrible to work from - please, someone give us studio remasters …
I found that best results was to setup your AVS script then run it through Vdub outputting to a lossless AVI file. Creates a huge 80-100gig file, but Topaz seems to love this way of doing things and I generally get a better result.
I have played with DS9 and in Vdub I correct Aspect ratio to proper 4:3 and in the AVS have a color correction part that works for all sources.
Currently I have moved this pipeline to a custom program (Delphi) that I have been working on to do heaps of ffmpeg stuffs. So now just drag file/s into here and in one press of button go from source vob to finished encode, thanks to everything being command line scriptable, including Topaz. Even creates the DGIndex file and AVS script needed.
1 Like
Unfortunately, it seems to be a rumor - I have been researching in that direction and I have barely found anything on this, the only other thing I ever found was another article that mentioned this, but there it was said that it would only be “upgraded/upscaled” by a studio (if anything).
But there are good news on my front - I have now three projects on-going (Upscaling Deep Space Nine and Voyager) as well as doing a sort of “Fusion” of Babylon 5 (meaning taking the 16:9 live action from the DVDs and the 4:3 CGI/VFX from the 1080p Web Remasters, but I am stretching the effects to conform to 16:9 without zoom and crop) - I will post something on my topic soon.
Oh yes, I know not all B5-Fans will like the stretch, but it’s something I can live with.
1 Like
Why not extend the 4:3 frame of the 1080p remasters with the 16:9 edges of the DVDs?
Also, your decision to stretch the VFX makes no sense at all. Just leave black borders, it’ll be less distracting.
As for extending with the edges from the DVD: That wouldn’t work, because the DVD version was zoomed in and cropped, so there would be visible differences.
Artificially painting in the missing edges also does not work (I have tried several software solutions and none so far was successful enough to be a viable option).
As for the general decision to stretch: I mentioned before, that many people would probably not agree on that and that is ok, it’s just my way of looking at things. I am not going to force anyone to like it. Just wanting to show what could be an alternative besides the DVDs in 16:9 and the 1080p 4:3 remasters.
That’s not weird, it’s perfectly sensible in my book. 
Star Trek is supposed to be a trek through the stars, the clue is in the name. DS9 is a static space station, it should be disqualified on those grounds alone. 
Perhaps it should be a separate franchise “Wormhole Camp: Deep Space Nine” 
Personally, I think every Trek-series has a right and reason to be there. It just shows how versatile the franchise is. There is something appealing for everyone.
I am not one of those who thinks like: If I like this version of the franchise, I have to dislike another. My taste actually changes from time to time. I was aware of DS9 during my youth but did not watch it the way I watched let’s say TNG or VOY. I grew up witth TOS reruns. In my 20’s DS9 grew on me a lot. These days, my favorites are DS9 and Babylon 5 (outside the Trek-franchise).
Why? Because I like the “realism” in the dynamics between the characters. After getting into the workings and reality of having a job, managing day-to-day life and how people act towards each other socially, my experiences come closer to DS9 and B5 (like “Why is he/she doing that? What’s the agenda behind that? How do I get the upper hand?”) instead of TNG (like “Let’s all be friends and talk about it”).
Not that I am that way, but I certainly had to cope with enough people acting like that, which pretty much took my innocence in that area.
It’s still something I regret.
But then again, when I need hope, the other Treks can build you up again. So I’ll watch that if it’s needed.
Hello there.
I just finished long and hard work, upscaling both Star Trek Deep Space Nine (Pilotepisode) and Babylon 5 (Episode 1 of Season 1).
First of, my samples on Star Trek Deep Space Nine (password is always “DS9Up”, without quotation marks):
Comparison Images (DVD on the left, upscale on the right):
Videosample (downsized from 4K to 1080p for size reasons):
Second, here are my samples from Babylon 5 (my approach here is to take the 16:9 live action shots from the DVDs and replace the CGI/VFX shots with the 4:3 HD remastered versions - those are stretched to 16:9 without cropping ; I know, not everyone will like that, just wanted to show an option here). The password here is always “B5Up” (without quotation marks):
Comparison Images (DVD on the left, upscale on the right):
Videosample (downsized from 4K to 1080p for size reasons):
Have fun and let me know what you think.
1 Like
I’m actually new to the program and new to this kind of thing myself. I’m also working on doing some Star Trek and Babylon 5 myself. Hoping to do DS9 and Voyager both, as well as Babylon 5 and probably even Crusade. I’ve found that when I try to do any enhancements on DS9, Voyager or Babylon 5 through Topaz, the video looks good, but I have no sound. Have you encountered this issue? And if so, how did you get around it? I’ve done work on some other shows that all came from DVD’s, but they did all have sound afterwards. From what I’m hearing, I have to extract the audio and then merge that audio with the new video after the enhancement is done? That’s disappointing that Topaz can’t handle that as expensive a program as it is.
