I would like to point out main disadvantages of current barrel distortion solution know as "pixel mapping":• Black borders around image (21% of screen is not in use)• higher resolution (than screen) buffer slows down FPS• current distortion is post-effectNow that you know this mystery I can describe to you the solution that I invented.Basically lets see how 3D objects are projected on flat "screen"This is oversimplificated presentationTo create distorted image, projection on curved surface can be done, but you must know that computer sees single polygons as projected objects...and wall divided by 3 polys looks like this:but... we can cheat it!What is the difference between flat projection and curved projection?The distance between focal point and the pixel! :Dsimple presentation –>Moving focal point for every pixel creates curved image.How to achieve it?Assume that you have 1920x1080 screen and 110 fov:Xres = 1920Yres = 1080FOV_H = 110Distance between pixel and focal point for left/right middle edge pixel is:DistancePixToFocal = 960/ sin( FOV_H / 2 )Focal4ThisPix = sqrt( DistancePixToFocal**2 - DistanceFromCenter**2 )Focal points values can be calc to file and then readed when needed ;)Aspherical distortion for Oculus RiftGraph shows different projections for pixel distance.Changing DistancePixToFocal to the value from graph creates flowing effects:This graphs represent relationship between distance from pixel to center of the screen and pixel to focal point distanceIf you like it please leave the comment...I'm already making code for thisEDITI changed links to images ;)

This "new" method you invented has been a standard (commonly used for domed screens) for many years:https://developer.oculusvr.com/forums/viewtopic.php?f=33&t=965&p=11107#p11107Sharing your implementation is of great service to the VR community. Thanks!

As well as the distortion there is also the chromatic aberration which is harder to correct for with something other than a post processing effect, unless we used something like sub pixel rendering. Another advantage of post processing distortion is you can do timewarp at the same time. Hmm I wonder if it would be possible to create an achromatic lens and render with the same spherical distortion so the timewarp could be by shifting the image without any extra distortion steps.

"mrmonkeybat" wrote:As well as the distortion there is also the chromatic aberration which is harder to correct for with something other than a post processing effect, unless we used something like sub pixel rendering.If you ray-trace for distortion you can do it for the three different wavelengths as well to account for chromatic aberration. There is no need for a post processing effect.The difficulty in the ray-tracing approach is to come with the correct parameters for the aspheric surface lens profile and to implement intersection and normal calculations for aspheres which is not trivial (iterative solution, but can be pre-processed).

"Fredz" wrote:If you ray-trace for distortion you can do it for the three different wavelengths as well to account for chromatic aberration.I wanted to use a mesh to overcome the slowness of the Raspberry Pi rendering, and it was the chromatic aberration that convinced me that it wasn't possible. It isn't a matter of drawing three lines at different wavelengths.Red, Green and Blue smear out across multiple pixels with fractionally different smear amounts. Trying to duplicate the smear with lines would require drawing an "affected area" hundreds of pixels thick for every single pixel.So the approach of pixel shaders for an input texture mapped to an output texture is as much of a requirement as the aberration effect of the physical lenses themselves. I think Carmack already gave up on meshes?

"geekmaster" wrote:This "new" method you invented has been a standard (commonly used for domed screens) for many years:https://developer.oculusvr.com/forums/viewtopic.php?f=33&t=965&p=11107#p11107Sharing your implementation is of great service to the VR community. Thanks!You are missing the hole thing. This "moving Focal point" method is not a ray-tracing thing, its more like dividing the screen in tiny rings and using different camera settings for each. I think this link could misled you: , here is updated versionI'm already making it work in Maxon Cinema 4D as Python plugin."mrmonkeybat" wrote:As well as the distortion there is also the chromatic aberration which is harder to correct for with something other than a post processing effect, unless we used something like sub pixel rendering. Another advantage of post processing distortion is you can do timewarp at the same time. Hmm I wonder if it would be possible to create an achromatic lens and render with the same spherical distortion so the timewarp could be by shifting the image without any extra distortion steps.Yes, you have right. Best way for chromatic aberration is the post effect -pixel mapping. But it can by done better 8-)Currently RGBY channel shift is used, but real chromatic aberration (produced by OR-VR lens) has smooth transition between colors. By using Pixel-color-hue to shift-multiplier graph the program can read value of how much he must move each pixel to get perfect correction.

Barrel distortion NEW SOLUTION MADE!!! | Meta Community Forums

15 Replies

geekmaster
Protege
11 years ago
This "new" method you invented has been a standard (commonly used for domed screens) for many years:
https://developer.oculusvr.com/forums/viewtopic.php?f=33&t=965&p=11107#p11107

Sharing your implementation is of great service to the VR community. Thanks!
MrMonkeybat
Explorer
11 years ago
As well as the distortion there is also the chromatic aberration which is harder to correct for with something other than a post processing effect, unless we used something like sub pixel rendering. Another advantage of post processing distortion is you can do timewarp at the same time. Hmm I wonder if it would be possible to create an achromatic lens and render with the same spherical distortion so the timewarp could be by shifting the image without any extra distortion steps.
Fredz
Explorer
11 years ago
"mrmonkeybat" wrote:
As well as the distortion there is also the chromatic aberration which is harder to correct for with something other than a post processing effect, unless we used something like sub pixel rendering.

If you ray-trace for distortion you can do it for the three different wavelengths as well to account for chromatic aberration. There is no need for a post processing effect.

The difficulty in the ray-tracing approach is to come with the correct parameters for the aspheric surface lens profile and to implement intersection and normal calculations for aspheres which is not trivial (iterative solution, but can be pre-processed).
steve
Honored Guest
11 years ago
"Fredz" wrote:
If you ray-trace for distortion you can do it for the three different wavelengths as well to account for chromatic aberration.

I wanted to use a mesh to overcome the slowness of the Raspberry Pi rendering, and it was the chromatic aberration that convinced me that it wasn't possible. It isn't a matter of drawing three lines at different wavelengths.

Red, Green and Blue smear out across multiple pixels with fractionally different smear amounts. Trying to duplicate the smear with lines would require drawing an "affected area" hundreds of pixels thick for every single pixel.

So the approach of pixel shaders for an input texture mapped to an output texture is as much of a requirement as the aberration effect of the physical lenses themselves. I think Carmack already gave up on meshes?
JMF
Honored Guest
11 years ago
"geekmaster" wrote:
This "new" method you invented has been a standard (commonly used for domed screens) for many years:
https://developer.oculusvr.com/forums/viewtopic.php?f=33&t=965&p=11107#p11107

Sharing your implementation is of great service to the VR community. Thanks!

You are missing the hole thing. This "moving Focal point" method is not a ray-tracing thing, its more like dividing the screen in tiny rings and using different camera settings for each. I think this link could misled you:
, here is updated version

I'm already making it work in Maxon Cinema 4D as Python plugin.

"mrmonkeybat" wrote:
As well as the distortion there is also the chromatic aberration which is harder to correct for with something other than a post processing effect, unless we used something like sub pixel rendering. Another advantage of post processing distortion is you can do timewarp at the same time. Hmm I wonder if it would be possible to create an achromatic lens and render with the same spherical distortion so the timewarp could be by shifting the image without any extra distortion steps.

Yes, you have right. Best way for chromatic aberration is the post effect -pixel mapping. But it can by done better 8-)

Currently RGBY channel shift is used, but real chromatic aberration (produced by OR-VR lens) has smooth transition between colors. By using Pixel-color-hue to shift-multiplier graph the program can read value of how much he must move each pixel to get perfect correction.
2EyeGuy
Adventurer
11 years ago
There are only 3 possible pixel hues: red, green, and blue.
JMF
Honored Guest
11 years ago
"2EyeGuy" wrote:
There are only 3 possible pixel hues: red, green, and blue.

That's why graph is needed, if you have pixel in R:G:B and its 255:255:0 (basically yellow ;) ) or 128:128:0 (dark yellow) program can read offset value corresponding to this color on the graph. As you can see on this sample graph blue colored pixel would have more offset than yellow

Here is relationship between the "hue" of colors with maximal saturation in HSV and HSL with their corresponding RGB coordinates.

Chromatic aberration occurs because lenses made out of glass have different magnification "power" for each color hue.

Mapping pixels by their color hue is perfect solution for chromatic aberration correction, but with pincushion distortion correction its like double mapping. I think that making barrel distortion at the rendering time by maneuvering focal point is the fastest solution.
bluenote
Explorer
11 years ago
"JMF" wrote:

You are missing the hole thing. This "moving Focal point" method is not a ray-tracing thing, its more like dividing the screen in tiny rings and using different camera settings for each.

But how do you want to do that on fragment level? Typically you convert from world coordinates to clip coordinates in the vertex shader and the fragment shader just loops over all pixels of each triangle. Using a different focal point per pixel means that you can no longer just transform the vertices of a triangle globally. And varying the focal point does not result in an affine transformation, so there is not straightforward way to do that with rasterization. There is actually a paper which (iirc) also approaches distortion on the transformation level (so it is similar to your varying focal point) and tries to solve the issues of the non-affine transformation by GPU based tesselation (better read it, I'm not sure if I remember that correctly).
AntDX316
Honored Guest
11 years ago
the best way for DK2 to work with all games is the ability to have NATIVE DK2 head tracking and correct image mirroring in ALL games

the foundation is there its just the way the camera should be for DK2 to work, like the track doesn't exist : (

it would be so simple if people really cared enough

its like trying to find aftermarket Daewoo Leganza parts but only a few exist like the lowering springs but not really anything else where as Honda Civics, you can find parts for EVERYTHING and it's so cheap
JMF
Honored Guest
11 years ago
"bluenote" wrote:
"JMF" wrote:

You are missing the hole thing. This "moving Focal point" method is not a ray-tracing thing, its more like dividing the screen in tiny rings and using different camera settings for each.

But how do you want to do that on fragment level? Typically you convert from world coordinates to clip coordinates in the vertex shader and the fragment shader just loops over all pixels of each triangle. Using a different focal point per pixel means that you can no longer just transform the vertices of a triangle globally. And varying the focal point does not result in an affine transformation, so there is not straightforward way to do that with rasterization. There is actually a paper which (iirc) also approaches distortion on the transformation level (so it is similar to your varying focal point) and tries to solve the issues of the non-affine transformation by GPU based tesselation (better read it, I'm not sure if I remember that correctly).

You pointed right thing. This method would work only in ray tracing :(
But I found interesting publication on this site
Here are some pics