DaVinci Resolve

misoversaturated

misoversaturated

0
After I read that current TV sets (don't own one) can scale up movies and double the frame rate in real time and good quality too I was looking for a software doing this on a Mac and found DaVinci Resolve, a video editor by the camera maker Blackmagic, which is freely available on the Mac App Store.
The quality of scaling ("super scale") as well as retiming ("optical flow") is really good and works well with 3D animations.
After testing with Blender I just applied it on an old Falcon animation from 2017, blowing it up from 1280x800@20s to 2560x1600@40s and I think it works rather well.
- The flickering in the grass due to caustic noise from blurred transparency is actually worse in the original file.
- Dropbox player is evil, download and view in VLC or Quicktime
- The link will expire later this year because my dropbox limit is near ;)

meadow.jpg

This means theoretically for a Full HD animation of 1920x1080@30fps one needs to render only 960x540@15fps which is eight times faster.
In reality there will probably all kinds of issues but it is worth to experiment with.

I was not able to get quicktime export working yet, no matter what color profile I choose the result is always too dark.
But tiff export works well and I'm used to batch denoise and grade animation image files in Photoshop anyways so it's not a big deal.
Maybe there are DaVinci Resolvers in our forum who could drop some hints?
 
Last edited:
Thanks for the info! I've heard of Resolve, but never really paid attention to it much. I know that it's really capable though for what it does. I'l have to check it out. Especially now that you've mentioned that it can upscale a video really well. Could save a lot of render time for HD videos.
 
You’re doubling your framerate too? Is it blending frames - or interpolating the changes? How does the motion look? (Sorry, not at my Mac right now)
—shift studio.
P.s. on a similar note, Topaz AI Gigapixel is a relatively new up-resing application that is superior in general to all the older blow-up software. Render at 1/4 res, then blow up 200% - looks quite good in many cases and takes 1/4 the time.
 
Shift Studio said:
You’re doubling your framerate too? Is it blending frames - or interpolating the changes? How does the motion look?
Click to expand...
Yes, interpolating with few artefacts (see linked video) and resize in one go - Metal accelerated on the Mac.
Thanks for the Topaz tip, I'll check that out (always in need for better resize tools)!
 
Thats an impressive scene and nice camera work too.
The video looks good. playing at full 30fps, fast panning motion near camera is a little jumpy between 10s - 20s. never saw the 15fps version though.
Thanks for sharing it, and the info too.
--shift studio.
 
Shift Studio said:
Render at 1/4 res, then blow up 200% - looks quite good in many cases and takes 1/4 the time.
Click to expand...

Well, I don't know how long topaz takes to blow up the pic but 1/4 res should render in a 16th of a time at maximum (if you take in account some stuff like displacement, it can be significantly more and then well be worth it). So if I get you right, topaz needs approx. the same time for blowing up the pic as the rendering?

I'll keep an eye on it. Do you have an example or two of what it really can do with a pic?
 
@ Hasdrubal- by 1/4 res, I meant 1/4 the total number of pixels. But 50% might have been more clear. i.e. 1/2 width and 1/2 height. Follow me?
AI Gigapixel’s process times are relatively small almost negligible compared to render times.
Sure, I’ll render two images as described and post them.
—shift studio.
 
I modified the C3D Pig scene to get something with a fair amount of details to examine - I rendered to 500 spp, with Falcon at 2k and 4k.
Here are the results for this experiment:
Render-times-and-sizes.jpg

You can see that the render time is approximately 4x for the 4k render.
I used Topaz A.I. Gigapixel v2.1.1 to Enlarge the 2k image up to 4k. .
I enlarged with 'Reduce Noise and Blur' set to 'None' and also set to 'Moderate'. Both took 48 seconds.

The resulting enlargements are good, but not exactly the same as the 4k render.
Some image areas are slightly less sharp/detailed and other areas are more sharp/detailed than the reference 4k image.
Some image areas where white quickly transitions to black are too sharp and could use some anti-aliasing.
Are the results good enough? I think that the answer is 'yes' in some cases, and 'no' in others.
Cutting render times down to ~25% is attractive though!

In my opinion, its not appropriate to evaluate this kind of thing in a web-browser, as you need to view at 'actual pixels' (100% zoom).
So I'm including a dropbox link to download and evaluate yourself.
I've include a layered file so you can easily compare to the reference 4k render.
https://www.dropbox.com/s/ylib1l308t67kh8/AI Gigapixel on C3D renders.zip?dl=0

I'm curious to see what you guys think.
@misoversaturated, sorry for hi-jacking the thread...

--shift studio.

edit: P.S. in near future, if anyone wants me to use AI Gigapixel on an image of theirs for test purposes, PM me.
 
Last edited:
@shiftstudio: Very impressive, thanks much for your effort!
That's exactly what I was hoping for: Users gather and share their ideas and workflows regarding resizing renders and retiming animations.
I should have chosen a more general thread title though.

While DaVinci is free, Topaz is about 100 bucks, right?
Probably the resize quality will be better, would be interesting to see an upscaled animation.
 
misoversaturated said:
@shiftstudio: Very impressive, thanks much for your effort!
That's exactly what I was hoping for: Users gather and share their ideas and workflows regarding resizing renders and retiming animations.
I should have chosen a more general thread title though.

While DaVinci is free, Topaz is about 100 bucks, right?
Probably the resize quality will be better, would be interesting to see an upscaled animation.
Click to expand...

Thanks Miso - glad it was of interest. Topaz AI Gigapixel is about $100USD. I believe I got it on promo for about $70.
I should have mentioned it doesn't do video... it can batch process a whole bunch of .pngs though - that could be a start.
Also I should mention its not so fast I read if you don't have a supported GPU.
Also, there is a problem with 'halos' on transparent bkgd images - but it its being worked on.
There is discussion about it here >> http://discuss.topazlabs.com/t/a-i-...png-files-with-transparency-alpha-mask/8779/8

--shift studio.
 
The white halos on images with alpha channels is a pretty common problem. It has to do with the alpha channel being pre-multiplied or not.
 
Swizl said:
The white halos on images with alpha channels is a pretty common problem. It has to do with the alpha channel being pre-multiplied or not.
Click to expand...

Thanks Swizl - yes that could be part of the problem - the thing is, if you look at this image, there is both white and black halo, and there is no black ‘background’. Normally you don't pre-multiply semi-transparent pixels with both white and black.
Gigapixel-Halo.jpg

regardless, I hope they fix it to have no pre-multiplying.

--shift studio.
 
Last edited:
Doh! That does look like a problem. :eek:

Hopefully they do fix that. It almost looks like a drop shadow applied. Very odd.
 
Shift Studio, thank you very much for your effort and your time. Especially that you made all files available is very helpful.

Actually the result is much better than I thought, even if the technique still has some problems. As this is fairly new stuff, a few problems have to be expected anyway. (I just don't remember exactly what it was, but some adobe program produced also such halos in files with transparent background in one version. So I'm sure they can manage that problem sooner or later).

I'm quiet impressed. Have up to now never seen anything better blown up at a 100 %. Yes, there are those problems you mentioned, and some artefacts (like a line for example) in the reflections and a few very coarse parts (like left of the ear in the right or the structure of the carpet reflection wherever that's visible). But on the other hand I can't imagine many things that are as hard as this glass.

I think it's good enough for many things. Could I use it at this stage? I'm tempted, but the answer at the moment is no, as I have to go for quality over time, especially as it seems necessary to edit a bit (which is my time, while render time is just the computer crunching bits away).

We shouldn't forget that we actually misuse this things. They are created for real photographs. In the past, though, I really had some projects where actual digital photos were quiet old and therefore to small, the quality as bad, that you couldn't even use it in the original resolution (in one case I used days to work on it. It looked great in the hand - at half the resolution which wasn't big enough to begin with). So if I could ever come to a similar situation, I'll happily buy.

In the near future, a few versions later, I'm sure it will be even better, the A.I. better trained, so it's quiet possible that I will use it even for 3d.

But again: Thanks a lot, shift. This really offered some insight.
 
Shift Studio said:
I modified the C3D Pig scene to get something with a fair amount of details to examine - I rendered to 500 spp, with Falcon at 2k and 4k.
Here are the results for this experiment:
View attachment 33838
You can see that the render time is approximately 4x for the 4k render.
I used Topaz A.I. Gigapixel v2.1.1 to Enlarge the 2k image up to 4k. .
I enlarged with 'Reduce Noise and Blur' set to 'None' and also set to 'Moderate'. Both took 48 seconds.

The resulting enlargements are good, but not exactly the same as the 4k render.
Some image areas are slightly less sharp/detailed and other areas are more sharp/detailed than the reference 4k image.
Some image areas where white quickly transitions to black are too sharp and could use some anti-aliasing.
Are the results good enough? I think that the answer is 'yes' in some cases, and 'no' in others.
Cutting render times down to ~25% is attractive though!

In my opinion, its not appropriate to evaluate this kind of thing in a web-browser, as you need to view at 'actual pixels' (100% zoom).
So I'm including a dropbox link to download and evaluate yourself.
I've include a layered file so you can easily compare to the reference 4k render.
https://www.dropbox.com/s/ylib1l308t67kh8/AI Gigapixel on C3D renders.zip?dl=0

I'm curious to see what you guys think.
@misoversaturated, sorry for hi-jacking the thread...

--shift studio.

edit: P.S. in near future, if anyone wants me to use AI Gigapixel on an image of theirs for test purposes, PM me.
Click to expand...


When antialiasing the crisp edges, you might find this video helpful: Enhancing 3D Renders #01

It's one of several video tutorials using Affinity Photo on 3D renders. Affinity isn't free, but it's cheap and cross platform and frequently as powerful as photoshop - at least for folk who don't need photoshop full time.

The master video tutorial page is here and it contains dedicated sections on useful skills for working with 3D renders and 32 bit images. I've worked with Adobe tools enough in the past to know that these techniques can easily be translated for use in Photoshop.
 
@MonkeyT: Good tip, no doubt, but shift studio's pictures come out of Topaz without any further work done by him. If he had, we wouldn't see much of a difference anymore. It's just about the possibilities of this quiet awesome app. In this example the mentioned anti-aliasing should come automatically from Topaz.

By the way, adobe worked in the last versions of photoshop on enlarging pictures, but the results are not even in the same ball park compared to Topaz, which is not unexpected. You can't blow up digital pictures very much because the detail just isn't there and it can't be computed. There is plain no way to it (which is why it's laughable when they in crime movies blow up pictures from bad surveillance cams several 100 percent, so they can read some text or whatever). There are no algorithms even thinkable that could reliable generate the missing information. It can't be done the old way.

The speciality of Topaz is it's A.I., which doesn't just try to connect some dots. It does create the missing detail (as the example show not always) with 'experience' gained from analyzing millions of other photos. Fox example, it knows how bork looks, so it recognizes bork on the original picture, then does enhance it with guessing what kind of detail is there in all the other similar cases and then computes information that's fitting to the original picture. And it is amazing, but in all probability nothing to what will be possible in 20 years from now (well, at least as long as humanity doesn't bomb itself back to the stoneage or some similiar results of N.S. (natural stupidity instead artificial intelligence ...). As soon as they stopped to rebuild human thinking the got finally results in this more than 50 year old research (there still is nothing around that could pass the turing test. On the other hand, over the years I met quiet a few people who wouldn't pass the test).

At the moment nobody can even guess what in some decades will be possible for A.I., especially combined with new hardware like quantum computers. Don't get me wrong, I'm all for it, but in different ways it's frightening (and dangerous, as there are already systems that 'detect' potential threats in crowds. Or some that analyze data to guess what you will do in certain scenarios or what you think about something (and this could lead to systems that will know which informations to give as in which ways to make sure that we'll do the right things). So of course this new technology will be misused in ways I can't even imagine (as it was with every one of them in the past), and the people will have to look out for future dangers stemming from it. So all this stuff like topaz is much more than new playthings for us or ways of doing or work faster. It's a glimpse into the future and therefore helps us understanding what's coming. Sorry, about digressing, but ...
 
Hasdrubal said:
@MonkeyT: Good tip, no doubt, but shift studio's pictures come out of Topaz without any further work done by him. If he had, we wouldn't see much of a difference anymore. It's just about the possibilities of this quiet awesome app. In this example the mentioned anti-aliasing should come automatically from Topaz...
Click to expand...

Understood. I wasn't specifically trying to resolve the original question, merely adding a references to the thread of a nice set of photo-editing techniques which were designed specifically to work with common (but subtle) issues in pure 3D rendering. I mentioned them because I know from experience that trying to achieve that last few percentages of quality in both photo editing and 3D can easily eat up a disproportionate chunk of your budget and your time. I worked in print, where recognizing "good enough" was a crucial production skill.
 
@MonkeyT : Actually those things would be well worth their own thread (actually even their own forum). You can find a ton of tutorials about that, but most are worthless because they are to heavy handed and / or take to much time. Sometimes I'm really at a loss what to do even while knowing how I could do it. The video you show I like because it's subtle.

The post-processing is something I could use more training :)
 
@Hasdrubal - I'm glad you found that exercise useful, and your insights are much appreciated.
@MonkeyT - thanks for your links and insights too. I still need to check the second one.
Also your comment
"... where recognizing "good enough" was a crucial production skill.:
Click to expand...
is so true – but a skill I still struggle with after all these years.

Hopefully we'll see some interesting improvements in the tech mentioned above in the years to come.
--Shift Studio.
 
You must log in or register to reply here.
Back
Top