Folks:  Needing some advice on a mixed aspect ratio project. I am working with decade-old video shot 4:3 and newer 4K media shot 16:9. The final project will be 16:9 so I have no problem with the newer footage but ideally we’d like to avoid the black sidebars when integrating the 4:3 footage – but that means pushing in enough so that the 4:3 left/right edges meet the wider 16:9 aspect ratio (and then also cropping top/bottom of the image).

The biggest problem we’re seeing is the noticeable degradation of the 4:3 image when we push in as much as we need to. Is there any way to maintain decent quality if I’m zooming in as much as 20-25% on the image? I’ve played a little with programs like Fonepaw Video Converter but it doesn’t seem like it does much (and aren’t these things just line-doubling)?

The majority of the material we’re using is in the 4:3 ratio – what does any of you recommend to maintain the highest image quality in such a situation?

Thanks, as always!

Tom

Tom, if you cannot stand to have pillar boxes on the sides (some people use graphic backgrounds or blur out and darken the same image on top of itself to fill the frame), then your best bet would be a hardware up-onverter. Some software programs work better than others. You could try DaVinci Resolve or After Effects. But going from 720x486 to 3840x2160 is a lot to ask of any software. Again, hardware options would give you the best results. Depending on where you live, some post production facilities would be able to do what you are talking about. I hope this helps.

Marc

And I will say that even hardware won't be that good.  SD to UHD is big ask.

I personally am really pushing for people to accept pillarbox.  But even that is a big ask for  SD to UHD.

Jef

One thing to try is to play around with Fusion from BMD.  The free version is fine for this.

The problem is the VERY steep learning curve if you have not used a node based compositing tool.  The reason to try is access to a whole set of tools not available in most other tools.  A large choice of fliters which are really important when scaling.

Jef

EDIT: I should have been more clear.  The idea is to use Fusion to do upconverts to individual shots.  This gives you a high degree of control.  But it is not great for churning through a lot of material.  And dealing with sound is a big pain.  It is really designed to composite imagery.

Thanks to everyone for such fast replies. We've experimented with graphics on the sides but the director isn't crazy about it. If we can't get a decent upconversion (enough to make it look close to HD - we're not trying to make it look like the 4K...), we'll stick with the back.

I'll try the fixes yu've all sugegsted and get back just to chime in with success or failure. Thanks again.

Hi all,

I have been researching a bit in the past couple of weeks on upscaling solutions from SD to HD. I came across this article on RedsharkNews:

https://www.redsharknews.com/production/item/6227-how-ai-is-helping-to-make-the-best-classic-tv-series-restorations

in the comment section a user describes his workflow for using this software for using it on video. I've done some experiments with it on upscaling SD 16:9 interlaced to 1080p50.

The software Gigapixel had a new update yesterday which improved the quality quite a bit. There is a full functioning 30 days test version without watermark. However it's quite GPU intensive work. I have an GTX 970 and it takes about 1 1/2 hours to process 1 min of video (3000 frames)

My workflow for upscaling 625i 16:9 to 1080p50 is the following:

1. After effects

720x576 footage in 1024x576 timeline (to get square pixels). With a scanlines similar to this process:

http://avidscreencast.com/2009/11/04-changing-field-order/ to blend the 50 intelaced fields into 50 progressive. However that's still work in progress, haven't found the best solution yet.

--> export to 1024x576 TIFF images

2. Gigapixel

upscale TIFF images to 1920x1080 with "Suppress Noise" and "Remove Blur" settings to "High"

--> export as TIFF

3. DaVinci Resolve

import TIFF images as 1080p50 image sequence. Maybe another pass of noise reduction and basic color correction

--> export to desired format

I'm looking into this solution for upscaling a whole SD archive of a TV-station. From a current perspective it seems to computing intensive for applying it to a whole archive since it's not an automated process and it just takes to long on my current PC. Getting a latest graphics card would increase the process quite a bit as well as applying less noise reduction.

I have installed two graphics card in my system (for work in Resolve) but Gigapixel seems to ulitize only the main graphic card.

Furthermore I haven't found a nice solution for nicely converting 50 interlaced frames to 50 progressive frames. One option would be to simply deinterlace but I don't want to just discard half the information of the source clips.

But for someone looking for a solution to upscale just a few minutes of footage to implement it into a higher resolution format I hope this can help. It probably gives also good results of upscaling HD to UHD.

Maybe we can keep this thread a bit updated with forther solutions and insights.

Thanks,

Jojo

Great info, Jojo, thanks! I will also give your process(es) a shot and report back asap.

Tom

I've got quite a lot of SD material that I'm also trying to upscale. Currently I'm trialing a product called BlowUp 3, from Alien Skin software. This is an image vectorization plug-in for Photoshop/Lightroom (though it can also be used as a standalone app.). Basically it analyzes an image and turns it from a raster format to a vector format, at which you can resize it. It works pretty well.

My problem is dealing with the interlaced fields in SD. I still don't have a good workflow. At the moment, I'm trying to get a reasonably clean image out of MC using a timewarp set to Blended Interpolation into an image sequence, batch process the image sequence through BlowUp 3 to HD, importing that back to MC and then using Boris Red to add a small amount of gaussian blur to get rid of residual interlace artifacts followed by an unsharp mask.

The result is a little blurrier than I would like, but I do get clean edges. It doesn't help that most of my material was shot in relatively low light and is grainy to begin with. If the material was prpogressive, I'd get much better results.

Folks:

In looking briefly at GigaPixel and BlowUp, they seem to be software for still images, yes? While I know vidoe is really just 30 (or 29.97, or 60, or whatever your rate) stills per second, it seems it would take FOREVER to do any serious length of video material frame-by-frame (or am I missing a functiobn of thewse programs?).

Ah, now that I ac tually RUN Gigapixel it has an opening screening remidner that video IS just a series of still images... dh.

BlowUp 3 is quite fast (relatively speaking) It's probably about 2x realtime to enlarge a video sequence. But it does have a limitation for batch when used as a stand-alone application (I end up doing about 10 seconds at a time). It can be automated through Photoshop, but I'm sure that will take longer.

Let's face it though; there is no perfect way to enlarge a standard definition image to HD. First of all, most SD material is interlaced, and HD is progressive. The difference in movement with successive fields leads to artifacts. Even motion adaptive algorithms which attempt determine the horizontal difference between fields can't completely compensate. Frame blending results in a softer image. Interpolation tends to lead to artifacts. There's always going to be something somewhere that you can't fix automatically.

It's much easier with progressive SD. This has a lot fewer artifacts when enlarged.

I'm facing the issue now that when I run GigaPixel AI, even for a single frame modification, I get the Windows Display Driver timeout error. I have played with the registry key and maxxed it out and it sems even a shorter time now that the driver crashes... I've got an Nvideo Quatro 2200 which has (thus far) been up to everything I've asked of it... anybody know how to override the display driver (would actually turining OFF the display matter (I'm assuming not)?

Tom

My own reply - when you use the default settings with GigaPixel it allows a preview screen. I minimized this and it seems to help - I was able to tcode ten or so before a crash...

Yes, it's a still image only program, hence the conversion to TIFFs. And yes, it takes forever. The time reference I gave for one minute was with an older version of Gigapixel. With the recent version and all the settings to highest quality 1 minute of SD footage (50p --> 3000 images) now takes 8 hours to render ... :-)

I have a GTX 970, which is already a bit outdated. A Quadro 2200 is quite a bit less capable. To bring the render times in some reasonable form I think one need to be looking at the top end of the graphic card market, GTX 2080.

If you have troubles with the graphic card driver in Gigapixel you can go to the preferences and change the processing mode to CPU. But I think it might be quite a bit slower, haven't tried it yet.