- Tue Jun 20, 2006 7:11 pm
#162928
Here comes a short comparison of a well-known V1-rendering provided by Nextlimit and a real-life photo. The purpose of my comparison is not to complain about anything, but to be able to get a grasp on the reason for the unsatisfactory look of V1-renders and maybe to contribute something to a betterment.
I converted the Cornell Boxes to greyscale bitmaps because it was easier to measure the brightness-values. The numbers right of the picture below show the RGB-brightness-values and the ratio between them.
As you see, the ratio of the brightness-values differ more, the nearer the lit surface gets to the lightsource. It is obvious, that the falloff of light-intensity in V1-renderings is very different from reality. The ratios show, that the differences are not just random, but would give a quite constant changing curve if pictured graphically. So, tuning gamma and burn will never lead to a satisfying output as Mike Verta tried in this thread, because the problem has nothing to do with global image parameters but with internal calculations that occur before.
The next step I will take, is to compare light distribution and brightness-falloff, depending on the distance between lightsource and surfaces, between Beta and V1 to see if there are similar differences.
Any thoughts and contributions on my comparison?
Next round: Comparison between Beta 1.2.1 and V1.0
I built a simple box that has an emitter plane at the top and otherwise only diffuse materials with RGB 200/200/200.
The Beta shows a much more balanced light-distribution than V1. I only have measured the shadow areas and noted the difference in brightness right of the picture, because I have “exposed” the image to the mid-tones. There is much more tonal detail in lights and shadows in Beta than in V1. I think that’s what is disturbing to the human eye, which is very sensitive to such subtle differences.
I made a crop from the middle of both pictures to show how Beta and V1 behave in mid-range. Beta shows a much better contrast and tonal detail in the mid-range while V1 looks somewhat flat. When keeping the shadows in mind that became soaked black, this is quite a bad performance. If V1 had more overall contrast, it should give more contrast to the mid-tones as well. But it doesn’t. So it must have something to do with light-distribution. Not enough bounces maybe?
At the moment, I don’t find any simple conclusion like: “there’s too much contrast” or “light gets absorbed too fast”. I’m no physicist by the way, but a quite serious photographer for about a decade now. So, my photographers instincts just tell me: Something is off
next test: comparison of grey scale and light-falloff
I tested, if the perceived visual differences are caused by different types of gradation curves or if light falls-off different.
From my observation I’d say the first seems to be the cause. I built a scene with a simple greyscale parallel to the camera plane to have a measurement for gradation curves and lots of planes that have increasing distances from the camera to test the fall-off. The first comparison is taken using standard settings the second with different Burn and Gamma. Obviously it’s impossible to make the tone curves match each other at the whole spectrum this way. Just have a look for yourself.
Now, it would be VERY interesting to transfer this test to real world and compare perceived reality taken on film/image sensor with Maxwell.
I uploaded the MXS here.
I converted the Cornell Boxes to greyscale bitmaps because it was easier to measure the brightness-values. The numbers right of the picture below show the RGB-brightness-values and the ratio between them.
As you see, the ratio of the brightness-values differ more, the nearer the lit surface gets to the lightsource. It is obvious, that the falloff of light-intensity in V1-renderings is very different from reality. The ratios show, that the differences are not just random, but would give a quite constant changing curve if pictured graphically. So, tuning gamma and burn will never lead to a satisfying output as Mike Verta tried in this thread, because the problem has nothing to do with global image parameters but with internal calculations that occur before.
The next step I will take, is to compare light distribution and brightness-falloff, depending on the distance between lightsource and surfaces, between Beta and V1 to see if there are similar differences.
Any thoughts and contributions on my comparison?
Next round: Comparison between Beta 1.2.1 and V1.0
I built a simple box that has an emitter plane at the top and otherwise only diffuse materials with RGB 200/200/200.
The Beta shows a much more balanced light-distribution than V1. I only have measured the shadow areas and noted the difference in brightness right of the picture, because I have “exposed” the image to the mid-tones. There is much more tonal detail in lights and shadows in Beta than in V1. I think that’s what is disturbing to the human eye, which is very sensitive to such subtle differences.
I made a crop from the middle of both pictures to show how Beta and V1 behave in mid-range. Beta shows a much better contrast and tonal detail in the mid-range while V1 looks somewhat flat. When keeping the shadows in mind that became soaked black, this is quite a bad performance. If V1 had more overall contrast, it should give more contrast to the mid-tones as well. But it doesn’t. So it must have something to do with light-distribution. Not enough bounces maybe?
At the moment, I don’t find any simple conclusion like: “there’s too much contrast” or “light gets absorbed too fast”. I’m no physicist by the way, but a quite serious photographer for about a decade now. So, my photographers instincts just tell me: Something is off
next test: comparison of grey scale and light-falloff
I tested, if the perceived visual differences are caused by different types of gradation curves or if light falls-off different.
From my observation I’d say the first seems to be the cause. I built a scene with a simple greyscale parallel to the camera plane to have a measurement for gradation curves and lots of planes that have increasing distances from the camera to test the fall-off. The first comparison is taken using standard settings the second with different Burn and Gamma. Obviously it’s impossible to make the tone curves match each other at the whole spectrum this way. Just have a look for yourself.
Now, it would be VERY interesting to transfer this test to real world and compare perceived reality taken on film/image sensor with Maxwell.
I uploaded the MXS here.
Last edited by SJ on Thu Jun 22, 2006 11:47 pm, edited 5 times in total.
- By choo-chee
- By Andreas Hopf