Compression/Upscaling thoughts while animation is being rendered.

Hello.
First, I want to apologize for the lack of updates related to the ongoing project. I know it's bad, but I wasn't expecting it to take that long. I thought I would be able to finish it quickly, avoid making WIP posts and just do the final result.
I'll talk more about it in the next post.

I was meant to make a post about compression/upscaling earlier, but simply kept moving it further and further, avoiding it and being lazy. It's time to fix this.

About a month ago, I finished a project, uploaded it to the cloud, and picked a bad name for it, something like: "name_deep_compression.mp4".
At the time, I wasn't thinking it might sound confusing, but it actually was.



And he's right.
When you see the name like that, what's the first thing that pops in your mind? I immediately see DVD discs movies, 8 movies on each side and abysmal 360/480p quality.
But this isn't what's going on with the final files. Smaller file size might be misleading, they all use modern and advance compression methods, pushing the file size to the minimum and retaining most of the quality of the raw footage. It wasn't like the quality was terrible, it's just the name was poorly chosen.

Initially I wanted to talk a bit different metrics that are commonly used to compare original footage to compressed one, but I am not sure if anyone would be interested in it. Right now VMAF from Netflix is very popular and generally want to be above 80%. Then again, if you see artifacts like on the left (floor dents), you should try lighter compression, even if the metrics says it's fine.

What about 4k?

Well, old projects are done, but all the new ones will be rendered in 200% native resolution, retaining more details and keeping the original image intact. What does it mean intact?
So, there was a suggestion to simply upscale the footage and make it crispier. I tried it and I didn't like it. Sure, this made it much more crispy, but it made EVERYHING crispy, even things that shouldn't be like DOF, halation and many different effects. And I am sure you can play with different models, change the order of things you upscale and so on.
But reality is - this doesn't really solve any problems when it comes to things I can render myself. If I need finer details, I can set a higher resolution and get much better image quality than any upscaler can produce.

Some of the artifacts. You can see her eye got much crisper, but her skin details completely lost. You can see DOF being "fixed" as well as the camera lens effect almost removed.



Still, there is a use cases for them, this is a good example. The blanket had a very low res texture and looked like shit in a close up.

What about uncompressed?

No, simply no. It's not meant for real time playback. "Bedtime training" was about 90 seconds and raw data was 60GB 8bit and 120GB 10bit. You likely won't be able to play it on a PC, let alone on a phone. Also, where would I store it, on the cloud? I mean, sure, for how many animations? The sizes are ridiculous, stutters are insane and there is simply no point of doing it. You'll get much more detail from a new 200% resolution than from a raw files with an old one.

What's next?

Animation waiting room. So, it's 1150 frames, ~2 minutes for each frame. Should be ready in 30 hours if everything goes right, there are no more fixes and I don't need to return the GPU I borrowed.

Thank you for your patience and I'll see you in the next post!