Four Quadrant Rendering for Large Bryce files for Printing
Horo promted me to investigate this after our responses to a member post [ http://www.daz3d.com/forums/discussion/3962/ ].
The problem was:
(1) We wanted to render very large images for printing using only Bryce. The subject matter may be such that an image would take many days or even weeks to render. If the render time could be reduced to 50% or even to 70% then days of render time may be saved.
(2) We wanted to have curved perspective, which corresponds with reality and which moreover could be adjusted to any amount of curvature to suit a specific scene composition. In this way we would get good looking compositions and eliminate much of the distortion inherent in the straight line perspective of Bryce which is particularly noticeable when very wide angles of view are used.
(3) We did not want to have to stitch images together using special applications [because of the straight line perspective]. Although this method, quoted by Horo [ http://www.horo.ch/raytracing/tuts/online/tut21/minitut21_en.html ], is suitable for many applications, we wanted to find a simpler one which would be practical for most applications [but perhaps not for all].
At first i considered the problem unsolveable, but some instinct would not allow me to give up. So i spent all night thinking and dreaming about it!. Then i remembered the work of M.C. Escher. If we could produce curved perspective lines which naturally follow the spacial volume and lens geometry, then by using a 4-Quadrant subdivision and rendering method we might come close to our goal. I had already used the 4-Quadrant method for Bryce-ONLY modelling with good results so maybe it will work here.
What we needed was some kind of Bryce Lens to produce the virtual curved perspective lines. But a more efficient method would be to use a curved mirror [we then only have one reflection to worry about]. Refractive lenses would be render intensive but the single reflective surface might be practical.
So i began building an experimental setup. Please see Picture_01. It consists of a 2D grid of fine lines [made from cylinders], a camera facing backwards and a large sphere which is perfectly reflective [this could be squashed or stretched a little bit to get various curved perspectives]. These 3 objects are all in line and both the grid and the camera are linked to the sphere [which allows the setup to be moved and rotated as a set].
Then an object was made and placed on the other side of the grid but also very close to the grid. We then made a nice composition which rendered quickly. The intention was to remove the grid and change the materials before the main concurrent render with 4 instances of Bryce all running at once.
We still have to determine if the setup will work more universally. At present it works for a square scene with the camera level and horizontal. Please see Picture_02. Here you can see the 4 Quadrants and you can see there probably will be no distortion when we join the 4 pictures again later [ because of curve tangency were they cross the vertical and horizontal centre lines].
The procedure is as follows: First a Master File is made which can be loaded again and again. I used a test size of 800 x 800 pixels. Then we make the first Quadrant using a Plop-Render Marquee. This is rather tricky to do and requires good hand and eye coordination. Please see Picture_03. Then we click on the little triangle at the bottom of the attribute stack and select "Zoom to Selection" Then we save the transform as a new file called "Section_01".
We repeat the above for the other 3 Quadrants ["Section_02", "Section_03" and "Section_04"] always starting again from the centre point after having reloaded the Master File. Please see Picture_04 for Section_03 before proper materials were applied.
Later when we have loaded them again and applied Materials etc., we can render them very large by selecting a "Render Resolution" of 1 : 4.00 in "Document Setup". When the 4 quadrants are assembled again in Photoshop we could thus have a very large Image.
For our render test we did something more modest. If it works on a small scale it will work on a large scale. Most likely it will work with other aspect ratios than the square. Will try that out later.
When we have all our Segment Files we delete the grid in each of them. In this experiment we applied materials that would take some time to render but not too much. We only wanted to test the principle. I made the sky in the deep texture editor and it is applied to the Cumulus Clouds in the Sky Lab. The material for the MetaBall Object and for the Water Plane are by David Brinnen as provided with Bryce 7.1 Pro. The lighting is kept simple.
The complete scene [800 x 800] was rendered with "Aggressive Optimization" and antialiasing set at normal. The CPU was running at 50%. It took 52 minutes 30 seconds to render. Then the 4 Sections were loaded and the size set to 50% [400 x 400] and "Report Render Time" turned on. The four renders were then started quickly one after the other. We had 4 concurrent renders going and the CPU was running at 100%.
The results were as follows:
Section_01: 19 min 29 seconds
Section_02: 13 min 18 seconds
Section_03: 18 min 29 seconds
Section_04: 29 min 56 seconds
Comparing the full render time with the render time required by the most demanding of the 4 sections, we gained 22 minutes and 30 seconds. If it had been a very large image with volumetric effects and other high cost effects, we could have saved ourselves many days using this method. In addition we got a beautiful perspective unlike the usual ugly Bryce one. And of course this perspective is adjustable to anything we want for aesthetic purposes.
Next we assembled the 4 sections in Bryce using an array of 4 Composit Plates [Picture Objects]. This would normally be done in Photoshop, especially if we need to work on a very large image [or don't have much time] and perhaps also have a few imperfections to fix up. But for a small experiment like this we can do it in Bryce as well. The good news was that everything lined up perfectly! We got what we expected. Please see Picture_05.
Our conclusion [for now] is that the method is practical for very large files, but requires some skill marking out the Plop-Renders. It is perhaps to complicated to use for anything that renders overnight.
Next we need to try other aspect ratios and camera angles.