Lecture 13: Global Illumination & Path Tracing (9)
colinsteidtmann
Cornell boxes remind me of Drake's hotline bling.
GH-JamesD
Are there any significant differences between these two that I might not be noticing? To my eye, they look nearly exactly the same, but could there be any other differences between global illumination and the real world that aren't accounted for by the renderer?
Edge7481
@gh-jamesD The only thing I noticed are the edges are a bit more defined on the photograph compared to the render, but it could be unrelated to lighting
AlsonC
Really struggling to see the big differences here, besides the edges being a bit more defined on one of the photos, is there any other big things I should be looking at?
muuncakez
I think the goal is for there to be little differences. The harder it is to be able to tell the difference from the real photo, the better the unreal render. There is also a Cornell Box with 3 balls of more explicitly different BRDFs.
yangbright-2001
The rendering effect is so close to real, I even though the right hand size is more natural.. The overall color looks brighter
spegeerino
The only real differences I can tell is that the shadow lines in the photograph look slightly crisper, and overall the photo is slightly darker than the render. If I had to guess, the first issue is caused by some kind of overzealous antialiasing effect and the second is caused by an ambient light level that's slightly too high.
rcorona
In the graphics literature I'm curious what are the standard procedures for evaluating rendering methods and comparing them to each other?
I'm curious if there is a quantitative way of defining "ground truth" such that one could measure how realistic a rendering of a scene is.
Although I imagine one could measure the pixel-wise error between the real-life picture and the rendered image, I think there's factors that make it non-trivial, e.g. perhaps the rendering method is great but the materials specifications for the 3D scene aren't configured well and this leads to an "unrealistic" image.
DreekFire
I can't tell which one is real either - perhaps a much higher resolution image would allow us to tell. Ray tracing doesn't actually use ambient light - all lighting is actual rays coming from the light source. As far as evaluating rendering methods, I believe the typical method is to use a renderer which is known to be unbiased, rendering the scene with a huge amount of samples which takes a very long time, and using that as the ground truth. Then, different renderers can be compared by running them for shorter amounts of time and seeing which one matches the ground truth best. That way, even poorly specified materials will be accounted for.
Cornell boxes remind me of Drake's hotline bling.
Are there any significant differences between these two that I might not be noticing? To my eye, they look nearly exactly the same, but could there be any other differences between global illumination and the real world that aren't accounted for by the renderer?
@gh-jamesD The only thing I noticed are the edges are a bit more defined on the photograph compared to the render, but it could be unrelated to lighting
Really struggling to see the big differences here, besides the edges being a bit more defined on one of the photos, is there any other big things I should be looking at?
I think the goal is for there to be little differences. The harder it is to be able to tell the difference from the real photo, the better the unreal render. There is also a Cornell Box with 3 balls of more explicitly different BRDFs.
The rendering effect is so close to real, I even though the right hand size is more natural.. The overall color looks brighter
The only real differences I can tell is that the shadow lines in the photograph look slightly crisper, and overall the photo is slightly darker than the render. If I had to guess, the first issue is caused by some kind of overzealous antialiasing effect and the second is caused by an ambient light level that's slightly too high.
In the graphics literature I'm curious what are the standard procedures for evaluating rendering methods and comparing them to each other?
I'm curious if there is a quantitative way of defining "ground truth" such that one could measure how realistic a rendering of a scene is.
Although I imagine one could measure the pixel-wise error between the real-life picture and the rendered image, I think there's factors that make it non-trivial, e.g. perhaps the rendering method is great but the materials specifications for the 3D scene aren't configured well and this leads to an "unrealistic" image.
I can't tell which one is real either - perhaps a much higher resolution image would allow us to tell. Ray tracing doesn't actually use ambient light - all lighting is actual rays coming from the light source. As far as evaluating rendering methods, I believe the typical method is to use a renderer which is known to be unbiased, rendering the scene with a huge amount of samples which takes a very long time, and using that as the ground truth. Then, different renderers can be compared by running them for shorter amounts of time and seeing which one matches the ground truth best. That way, even poorly specified materials will be accounted for.