May 14, 2014 11:49 pm
There is a lot of talk about 8k in the Fulldome world at the moment. We recently presented a short talk on the subject at the IMERSA 2014 summit during the 8k night at Fiske Planetarium. We thought it would be helpful to make it available here to continue the discussion. We do plan to put a white paper together to formalise our research but for now here are our thoughts…
In 2011 we decided to convert our best selling dome film ASTRONAUT into Fulldome Stereoscopic 3D (FDS3D). We evaluated the possibility of doing it in 8k also. At the time with the computing power available to us it would have been impossible, especially alongside the R&D task of figuring out FDS3D production properly.
So 3 years on we have started evaluating 8k again, computers are faster and more 8k domes are emerging.
As a producer we need to ensure longevity of our product over it’s life cycle in the market place, plus we just love to be on the cutting edge of all things Fulldome!
We are finding that the right show can have a life span of 8 years and counting in the case of ASTRONAUT.
The technology of the dome continues to march on and our objective is to try to determine what the future proof gold master should be.
The world is not made up of pixels. Our visual system is made up of retinal cells, each capturing a tiny ray of light which is then transmitted as electrical signals to the brain where the image is then interpreted. Just as computer images are built from a matrix of pixels our visual system has a finite resolvability depending on the amount of retinal cells. As long as we can see pixels, we are seeing a stylised version of reality. The closer we get to this retinal resolution the more ‘real’ things seem.
Fulldome already has the wow factor. Most people who watch a Fulldome film at the National Space Centre have never seen one before. For those people, the sensation of tumbling through space is so impressive that it’s hard to believe what difference higher resolutions would have over the impact of immersion.
So what’s acceptable? What really matters to a general public? And what should we be striving for? We are not good test subjects to represent the average audience member. We are used to seeing high resolution images and have much higher expectations compared to an average dome visitor. But most of us can probably agree that watching anything below 1k on a dome is not the inspiring experience that watching 4k content is. So where do we draw the line? And do we settle for acceptable or do we push for exceptional?
Before we all punch the air and and dive head first into the world of 8k, let’s see what it actually means to produce an 8k film.
There’s more to 8k than just changing the render output settings to 8192×8192 and hitting render. Every asset in the scene requires a higher level of detail to hold up at this resolution. Polygon edges that aren’t visible at 4k can be visible and textures can look soft. There’s no point having extra pixels if they aren’t adding any detail or even worse, if they highlight the imperfections in the scene that weren’t visible at 4k.
Even when we have our 8k dome master and everything holds up at 8k, that image won’t be transferred to the dome without some extra processing and an inevitable reduction to the fidelity of the original image. In order to playback 8k video at 30fps or even 60fps, a lossy compression is normally employed to reduce the amount of data being pushed through the system. Lossy encoding can decimate large portions and although encoding algorithms are designed to retain detail, there is usually a reduction in perceivable detail caused by the process.
Here are some examples of geometry mapping and encoding artefacts
The flight of the pixel from the dome master to the screen via the playback system and projection system is a very difficult one. To truly achieve 8k on the dome at the end of this chain is very, very challenging.
This has to be part of the business case to create a 8k dome master. If the dome display systems cannot truly achieve 8k on the dome now or over the life cycle of your show then there is no point mastering at this resolution.
We often hear “4k content looks great on an 8k dome system”, ever wondered why that is? Because the “4k” dome systems aren’t always resolving true 4k on the dome and the current crop of 8k systems allow us to see 4k content natively as it was meant to be seen. The same is true for 8k systems, are we truly seeing 8k content on the current 8k systems? We have not spent enough time with 8k systems to say what the current crop are truly resolving. This is in no way an attack on anyone, we love the fact that the vendors are trying so hard to push the fidelity of the dome upwards but it is a frustration in our industry that there is no commonly recognised way to truly establish the spatial resolution of any given dome system to help us producers decide what we should be aiming for to maximise our budgets.
We would be happy to collaborate with others and produce test cards etc to create a more transparent understanding of the true resolution of dome systems out there.
So let’s focus on some initial findings to create true 8k content and leave the argument for true 8k projection for another time.
What’s the actual difference between 8k and 4k?
(Click image to download 8k frame)
Here we have a scene where all the assets are modelled and textured at a level that will stand up to 8k. You won’t see any low poly objects or blurry texture. The image is split up into vertical segments showing 8k, 4k bicubic, which has been upscaled to 8k using a basic method, and 4k with detail preserving upscaling, which is a more complex method of upscaling that should help retain sharpness. If you follow the white lines down you will see where the 8k and 4k images meet.
The difference is very obvious when viewed on a monitor at 1:1 resolution and the same should be the case on a true 8k dome display. This is by no means a great resolution test frame to really analyse how many pixels you are resolving on the dome but we feel it is a good example of a high-end, real-world shot. Also it is worth noting that the 8k render has not been sharpened so a bit more detail could also be squeezed out of it.
So how does this extra resolution impact render times. This shows the render times of each layer that make up the previous image
Accumatively we are looking at a jump of nearly 3 times from 4k to 8k in this example.
If we show the jump to 16k we are looking at a much steeper curve but we will come back to 16k later…
What takes 1 hour and a half to render at 4k, takes 4 hours at 8k and 19 hours to render at 16k. These results are dramatically less than the test result when rendering ASTRONAUT 3D 3 years ago.
This is due to increased render farm spec in CPU and RAM.
We see a similar challenges with file size.
This is a lot of data to manage and move around on a frame by frame basis. 4k 30fps content has a data rate of around 1.5GB/sec, 8k 30fps jumps up to 6GB/sec. When you consider the average SSD is capable of 500MB/sec read and even 10Gbit networks have a line speed of 1.25GB/sec and in a real world scenario more like 500-600MB/sec just reviewing these 8k frames is a significant challenge.
So the simple impact on production is more of everything!
After 8k, what’s next? Do we just continue to push resolutions higher and higher in an endless struggle against the pixel?
Well, no. As we mentioned before, our eyes have a finite resolution based on the amount of retina cells we have. For most people that resolution is generally considered to be 1 arc minute. that’s 1/60th of a degree.
This equates to 60 pixels per degree in digital terms. Anything beyond this is not resolvable.
So what is the retinal resolution of the dome?
So the retinal resolution of any dome if viewed from the centre is 6876 x 6876.
But… This is based on a viewpoint from the centre of the dome only. As the viewpoint nears the edge of the dome, the perceivable resolution must increase. To work out what resolution we would require for someone sat in a seat near the edge of a dome, we need to add some measurements to the equation.
So to figure out this retinal resolution for people sat up to 2 metres away from the screen of a 20m dome we need nearly 69k!!
If we figure this out for a few positions within the dome we can plot a graph and see how the retinal resolution increases exponentially as we near the edge of the dome.
So in summary…
True 8k content is achievable and we believe a worthy cause. Production costs are very steep. 8k provides near retinal resolution for most of the audience. 16k would provide most of the audience with retinal resolution but is currently out of reach. There needs to be transparency on what resolution the current and next generation of domes can truly resolve on the surface of the dome.
Higher resolution by itself is not a meaningful pursuit. It’s one area of film production which has an impact on the audience experience. All the other parts are just as, if not more important.
We hope you find this information helpful and we welcome your comments. Also please challenge our maths if you think it is wrong as we want to get this right to help inform our process, we are not preaching to anyone, we just want to make the best possible Fulldome content for audiences to enjoy for years to come and feel collectively as the Fulldome community we need to figure this out so budgets can be spent wisely and time spent on creating mind blowing experiences and not over rendering!
You can grab the original presentation we did at IMERSA here as a handy PDF >