Sign in to follow this  
Yours3!f

Raytracing artifacts

Recommended Posts

Yours3!f    1532

Hi,

 

I'm writing a cpu raytracer using my math library based on this tutorial:

http://www.flipcode.com/archives/Raytracing_Topics_Techniques-Part_1_Introduction.shtml

 

I got through it up to chapter 6, where I noticed that I get subtle artifacts in the rendering (see pic), which gets even more pronounced when I enable fast mode (only diffuse lighting, 3 fps).

 

Any idea what am I doing wrong?

 

download here:

https://docs.google.com/file/d/0B33Sh832pOdOaE1LU3ZqMWZBUzQ/edit?usp=sharing

 

EDIT: to run the exe on linux, set the LD_LIBRARY_PATH to ../external/lib, ../external/lib/tbb

you'll also need to create a symlink from external/lib/libGLEW.so to libGLEW.so.1.6 (for some reason it links to libGLEW.so.1.6 too... even though I have 1.7)

you'll also need to install gsl (GNU scientific lib)

you can also build it using cmake. To disable the tbb (Intel thread building blocks) and gsl dependencies, set in globals.h USE_QRNG to 0, and COUNT_INTERSECTIONS to 1. Make sure you're not linking to it after this. Note that this will disable threading and usage of quasi random numbers.

 

Best regards,

Yours3lf

Edited by Yours3!f

Share this post


Link to post
Share on other sites
Yours3!f    1532

I remember having similar problem, it was gone after I added multisampling;)

 

well I did think that that is the solution, but in the original sample the artifacts are not there even without multisampling.

 

btw did you create a similer gamedev thread?

Share this post


Link to post
Share on other sites
alvaro    21246
You have a bug. Using multisampling will only mask the problem. I would try to identify a pixel that gets assigned the wrong color and then step through the code, to see where the computations are going wrong. I have implemented a raytracer before and it didn't have this kind of issue.

Share this post


Link to post
Share on other sites
Tasaq    1530

btw did you create a similer gamedev thread?

Nope;)

And what Alvaro said is true that multisampling will only mask the problem, sorry for not mentioning that ;)

In my raytracer artifacts appeared in predictable places (like places with precision problem, for instance very far away from camera).

Share this post


Link to post
Share on other sites
GarFieldo    130

I have had a similar problem with my raytracer. When checking for intersections, I used to compare the factor for the direction with zero to determine whether the object is in front of the view plane:

...
t1 = object.intersect(ray);
if(t1 > 0 && ...)
...

It fixed once I compared to a value near zero, e.g. 0.000001 instead. I can only assume that some sort of floating point operation inaccuracy is the reason for the artifacts.

Edited by GarFieldo

Share this post


Link to post
Share on other sites
nfactorial    735

Artifacts like that can be caused by reflected vectors intersecting the object ifself, you can avoid them by moving the ray origin my a small epsilon ie. rayOrigin += rayDirection * EPSILON, or checking that the distance to the next intersected point is further away than the epsilon.

 

It could be something else too, but that would be my first guess looking at the picture.

 

n!

Edited by nfactorial

Share this post


Link to post
Share on other sites
Yours3!f    1532

You have a bug. Using multisampling will only mask the problem. I would try to identify a pixel that gets assigned the wrong color and then step through the code, to see where the computations are going wrong. I have implemented a raytracer before and it didn't have this kind of issue.

 I tried to, and fortunately the same pixels are wrong each frame, so I could debug it. I got to the conclusion yesterday that the shading is correct, however I think the intersections may be incorrect. (find_nearest() in main.cpp) I'll do further debugging, but I'm still learning the maths behind it...

Share this post


Link to post
Share on other sites
Yours3!f    1532

I have had a similar problem with my raytracer. When checking for intersections, I used to compare the factor for the direction with zero to determine whether the object is in front of the view plane:

...
t1 = object.intersect(ray);
if(t1 > 0 && ...)
...

It fixed once I compared to a value near zero, e.g. 0.000001 instead. I can only assume that some sort of floating point operation inaccuracy is the reason for the artifacts.

 

yeah I also noticed in the tutorials that it doesn't check for hitting something, but rather is it close enough?

Share this post


Link to post
Share on other sites
Yours3!f    1532

Artifacts like that can be caused by reflected vectors intersecting the object ifself, you can avoid them by moving the ray origin my a small epsilon ie. rayOrigin += rayDirection * EPSILON, or checking that the distance to the next intersected point is further away than the epsilon.

 

It could be something else too, but that would be my first guess looking at the picture.

 

n!

 

If you look at the code I'm already doing this:

main.cpp, 536:

result += color * vec3( p->mat.reflectivity ) * raytrace( ray( point + re * epsilon, re, ++curr_id ) ... );

Share this post


Link to post
Share on other sites
GarFieldo    130

Hm since this would be representative to 0.000001m (at least in my raytracer), this should be close enough.

I didn't notice that you already did the same thing by moving the ray away from the object as described by nfactorial (the result is the same as my approach). What is your value for epsilon?

Share this post


Link to post
Share on other sites
Yours3!f    1532

Hm since this would be representative to 0.000001m (at least in my raytracer), this should be close enough.

I didn't notice that you already did the same thing by moving the ray away from the object as described by nfactorial (the result is the same as my approach). What is your value for epsilon?

 

:)

globals.h, 24:

static const float fepsilon = 0.0001f;

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this