360° Dome Compositing
Whether you have been to a planetarium or have used any VR devices, you have seen what a 360° video looks like. It allows the audience to be looking at the sky, at the back, to the sides, as if the video is real-life vision.
Movie screens we traditionally use are rectangular. They are flat. Anything from 3D renders to 2D effects are mainly setup to come out as rectangular images so we can add one rectangle on top of the other and layer them up for any kind of visual effects we need for videos. Software kits compositors operate such as Nuke and After Effects rely on creating rectangular flat 2D images to work on. Recently Arnold started supporting full dome image rendering, meaning you don’t have to render panoramas and stitch them together to get a 360 video anymore; you can render CGI in full 360° and send them to your compositing software. Planetariums have been using the dome technology for many years now, but it can get time and money consuming to go back to your 3D renders to fix for everything that receives a note from the director. Instead, it can be much easier to bring your EXRs into Nuke and address the notes in comp. Now if the notes are simple creative additions to the color, you can easily use the CC tools inside Nuke without additional hassle (or AE, however since I am a Nuke Operator, I will be explaining the process through Nuke even though it should largely be applicable to AE.)
THIS CAN BE ACHIEVED BOTH WITH FOUNDRY’S OWN SPHERICAL TRANSFORM OR CARA VR’S C_SPHERICALTRANSFORM TOOLS
A quick search on some of the scientific papers on dome master and dome compositing articles will show you that there are not that many helpful practical tutorials out there and the ones that exist only talk about stitching scenes and making stereoscopic images. What the compositors need to find out is HOW TO USE NUKE’S RECTILINEAR TOOLS ON 360° RENDERS. Your search will at least tell you the idea behind how to manage using rectilinear effects on spherical images. The way to do it is to unwrap your image from spherical into a latlong image or a cubemap - in Cara VR tool set you could pick rectilinear. Once you find the exact settings to unwrap your spherical image into a rectilinear, you also have the potential to invert the settings and use the inverted settings on a rectilinear effect you want to use and turn that into spherical. ( This is an option you could easily use if you are wanting to get quick 2D effects such as glow or roto nodes to look spherical and over it on top of your spherical image)
The upside of turning your effects into spherical is that you don’t lose and quality from compressing your renders during the process of turning them into rectilinear or latlong maps from spherical versions. However, depending on your situation you might have to turn your VR video or 360° renders from CG into workable rectangles.
You do need to get specific details from your 3D artist about the format and 3D world location of your render to get an accurate representation of your spherical image in rectangular format. However, right off the bat there are a few things to keep in mind that will help you.
If you are using Nuke’s boxed in regular spherical transform node then you can choose your image input as 180 fisheye ( for half dome compositing) and out to cubic or LATLONG map- and again you might need help from the 3D department to guess the XYZ rotation inputs for your image.
If you are receiving a Latlong map from your CG artist to turn into spherical in Nuke, you need a 2:1 ratio, and you need to set your input settings accordingly. Latlongs always expect a 2:1 ratio.
If you are using Cara’s spherical transform node then make sure you are most likely receiving a fisheye render or a latlong so make sure to set your input correctly and your format ratio.
Fisheye is usually the lens rendered from renderers like Arnold and Equidistant is usually the default setting you want to use for estimated fisheye distortion. Equidistant is considered the most common and ideal model for camera solves and your renderer is most likely to use this model for 360° renders.
pan/tilt/shift is usually the easiest ways to navigate your sphere within the limited rectangle view you will end up with.
One of the biggest conundrums will be the cropping of your image/ filtering of it.
Best way to avoid losing your edges once you turn something into spherical from rectilinear again is a hack: Reformat your image to a format bigger right before your spherical transform math. Right after the spherical transform, you downsize your image into the original format. Because you will be stacking them right after each other you should be getting the least amount of filtering (only once because of concatenation math). And because Nuke will be applying your effects into the bigger format, you will lose less quality. The downside can be calculation time, but Cara VR really sets itself apart in this department because of its GPU calculation method, which is something the regular Nuke spherical Transform node does not pack as of now.
he eyes and a great go to font for titles, paragraphs & more.