In 194-26, we discussed how many rendered textures have the rendered/"video-game" look because they don't have the texture imperfections that will show up in real-world textures. The Surface Roughness texture seems to help alleviate this issue, but it's still somewhat prevalent (and even more so in the window in the back, which kind of looks like a painting).
jierui-cell
How do we get those textures at the very beginning? Are they generated by computer programs using some simulation / math equations? Or we first take photographs in real-life and then input them as texture to our programs?
Zc0in
There is no real 2D surface in real world. Adding some geometry to the twxture makes the image nuch more real. But will it be much more difficult to render couse it will add much more 'little reflection surface'. Or we can just add shade or something like this to use 2D to simulate 3D?
sZwX74
@jierui-cell For the example of this wood floor, we can get these textures by having a computer break down a photograph of the floor into something like color/distortion. We can also have the GPU computationally generate it based on previous textures it has stored. So in a sense, I think both ways that you mentioned are viable ways of generating an original texture.
Staffjamesfobrien
Do a web search on something like "wood shader" and you'll find some cool examples of procedurally generated wood. (Also lots of ads for wooden shades/blinds.)
austinapatel
There seem to be multiple ranges of textures from color, to bump maps, to surface geometry maps. Are these efficient mechanisms for storing these maps, or do they all need to be stored independently? I'm also curious to know if there are efficient methods for loading these textures since all of the textures will need to be loaded at once when rendering an object.
StaffDanCubed
I wonder if (and if so, when) we have reached the point where computer renderings can become 100% indistinguishable from real-world photos. I remember some years back it used to be the case that computers can generate very “good” image renders like this one, but it’ll still somehow look “fake” (just like this one) and just be slightly different to the point where a human can reliably identify them as generated. But nowadays it really seems like there are some renderings that are just impossible to tell apart. Like the ikea images mentioned a few lectures back, I probably would not be able to tell that they’re fake without someone informing me.
waleedlatif1
Beyond these three features of texture mapping (surface color, roughness, and geometry) are there any other features that can increase the realistic-ness of the photo for it to be indistinguishable from a real photo?
reinaw1012
In the three examples here, are they bitmap images placed on top of the floor model? Or are they functions that output a pixel value based on position/lighting/angle etc?
Staffjamesfong1
@waleedlatif1 Great question. For wood flooring, the 3 maps shown here are already pretty good.
But at Disney's Pixar, they use materials that have over 10 maps per surface! (See Figure 10 on this paper from their research group.)
We will get to talk about more complex materials in the material modeling section of this course.
Staffjamesfong1
@reinaw1012 In this case these are just plain ol' images. Mathematically, they are still "functions", even if they are just stored as pixel values instead of symbolic formulas.
People have also tried expressing materials without an explicit pixel map too. For example, this shader does not use any bitmaps.
In 194-26, we discussed how many rendered textures have the rendered/"video-game" look because they don't have the texture imperfections that will show up in real-world textures. The Surface Roughness texture seems to help alleviate this issue, but it's still somewhat prevalent (and even more so in the window in the back, which kind of looks like a painting).
How do we get those textures at the very beginning? Are they generated by computer programs using some simulation / math equations? Or we first take photographs in real-life and then input them as texture to our programs?
There is no real 2D surface in real world. Adding some geometry to the twxture makes the image nuch more real. But will it be much more difficult to render couse it will add much more 'little reflection surface'. Or we can just add shade or something like this to use 2D to simulate 3D?
@jierui-cell For the example of this wood floor, we can get these textures by having a computer break down a photograph of the floor into something like color/distortion. We can also have the GPU computationally generate it based on previous textures it has stored. So in a sense, I think both ways that you mentioned are viable ways of generating an original texture.
Do a web search on something like "wood shader" and you'll find some cool examples of procedurally generated wood. (Also lots of ads for wooden shades/blinds.)
There seem to be multiple ranges of textures from color, to bump maps, to surface geometry maps. Are these efficient mechanisms for storing these maps, or do they all need to be stored independently? I'm also curious to know if there are efficient methods for loading these textures since all of the textures will need to be loaded at once when rendering an object.
I wonder if (and if so, when) we have reached the point where computer renderings can become 100% indistinguishable from real-world photos. I remember some years back it used to be the case that computers can generate very “good” image renders like this one, but it’ll still somehow look “fake” (just like this one) and just be slightly different to the point where a human can reliably identify them as generated. But nowadays it really seems like there are some renderings that are just impossible to tell apart. Like the ikea images mentioned a few lectures back, I probably would not be able to tell that they’re fake without someone informing me.
Beyond these three features of texture mapping (surface color, roughness, and geometry) are there any other features that can increase the realistic-ness of the photo for it to be indistinguishable from a real photo?
In the three examples here, are they bitmap images placed on top of the floor model? Or are they functions that output a pixel value based on position/lighting/angle etc?
@waleedlatif1 Great question. For wood flooring, the 3 maps shown here are already pretty good.
But at Disney's Pixar, they use materials that have over 10 maps per surface! (See Figure 10 on this paper from their research group.)
We will get to talk about more complex materials in the material modeling section of this course.
@reinaw1012 In this case these are just plain ol' images. Mathematically, they are still "functions", even if they are just stored as pixel values instead of symbolic formulas.
People have also tried expressing materials without an explicit pixel map too. For example, this shader does not use any bitmaps.