Sign in to follow this  
Prune

OpenGL HDR multisampling AA looks poor for high contrast edges

Recommended Posts

Prune    224
I'm using 16x8 coverage multisampling (tried regular as well, with same results) in an OpenGL program, with 16 bit floating point textures and FBOs. The AA looks fine for edges that do not form very high contrast against their background. But if, for example, the edge is reflecting an bright area from the HDR environment map, and is against a background that is not very bright itself, the AA looks quite poor, almost as if there wasn't any. I'm guessing this has to do with the fact that tone mapping is done after the multisampling, but obviously that cannot be the other way around, unless I'm missing something... So, what's the best way to improve quality? Here's a screenshot, showing AA working OK except where the edge goes across the brightest part of the reflection:

Share this post


Link to post
Share on other sites
Sc4Freak    643
Yes, it's because you're doing tonemapping after MSAA.

I'm not sure about OpenGL, but in Direct3D10 custom multisample resolve allows you to do just that - multisample after you do your tonemapping. The feature should be available on all Direct3D10 compliant hardware - it's just a matter of whether or not OpenGL exposes this functionality.

Share this post


Link to post
Share on other sites
Prune    224
Is there a standard name for this? "custom mutlisample" or "custom multisampling" do not return many results in Google.

I should also note that with HDR textures the analogous problem exists, because the texture resampling is done before tone-mapping, and high contrast image edges make it look as if nearest neighbor was used instead of bilinear/mipmapped interpolation. Is there custom interpolation in Direct3D for textures as well?

I'm guessing that when I add bloom it would hide the poor AA a bit, since the bloom has most effect in exactly such high contrast areas, but that's kind of like trying to sweep the problem under the rug and is not a robust solution.

As a side question, should bloom be computed before or after tone-mappin? My guess is that the correct way would be that it should be before, since if it's done after, then two areas of the same size that are both saturated after tone-mapping, but have different true brightness, would end up having identical bloom amount, which is obviously wrong.

Share this post


Link to post
Share on other sites
Sc4Freak    643
Quote:
Original post by Prune
As a side question, should bloom be computed before or after tone-mappin? My guess is that the correct way would be that it should be before, since if it's done after, then two areas of the same size that are both saturated after tone-mapping, but have different true brightness, would end up having identical bloom amount, which is obviously wrong.

The point of tone-mapping is to adapt an HDR image down to LDR. I believe that bloom should be done before tonemapping, while still in HDR space. However, bloom shouldn't contribute to a scene's luminance.

Share this post


Link to post
Share on other sites
coderchris    304
Iv had this same exact problem. What i ended up doing was doing the tonemapping not as a post process, but at the end of my main material shaders. This way, the color gets tonemapped down to an acceptable range before actually rendering the sample, allowing the edges to be properly antialiased.

Another advantage to this is that you dont even need a floating point surface since the color is tonemapped anyway (assuming you saturate the tonemapped color). Unless of coarse you also want bloom, then you dont saturate and write out to a floating point surface

Share this post


Link to post
Share on other sites
Yann L    1802
A common problem. Even some offline renderers have this problem. Actually, multisampling after tonemapping is mathematically incorrect too (since the post-tonemap samples aren't linear anymore). But compared to the jaggies, it's usually more pleasing visually.

I have recently done some experimentation with a special adaptive tonemapper, that takes edge contrast into account. The idea is to detect steep interpolation gradients (as they are created by multisampling a HDR surface with high contrast) and decrease exposure along these edges. It works quite well, but I still get some annoying sparkling artifacts, probably due to lacking precision at the differencing pass.

Doing this isn't even that far fetched, since the human visual system does something similar in the real world.

Share this post


Link to post
Share on other sites
coderchris    304
I guess technically it wouldnt be correct, i hadnt actually thought about that until now; Though, I believe that the source engine does their tonemapping the way i described also; so either they have a special way to do transparency or its not noticable enough to tell :p

Actually, i can think of one way around it. WHen you render transparents, dont enable blending, but rather read the backbuffer and "un-tonemap" the color you read, then do your blending in the shader, tonemap the new color, and return that.
A drawback here is that if you want to draw one transparent on another, you have to do some render target changing...

Share this post


Link to post
Share on other sites
Prune    224
Quote:
Original post by coderchris
Iv had this same exact problem. What i ended up doing was doing the tonemapping not as a post process, but at the end of my main material shaders.

With this, you can then only use local tone-mapping operators.

Share this post


Link to post
Share on other sites
LeGreg    754
Quote:
Original post by Prune
Quote:
Original post by coderchris
Iv had this same exact problem. What i ended up doing was doing the tonemapping not as a post process, but at the end of my main material shaders.

With this, you can then only use local tone-mapping operators.


That's not true, Half Life 2 was doing just that : rendering everything to a LDR target, doing the tonemapping at the end of every shader. Their tone mapping operator was a global one (a simple division by a constant exposure value).

Quote:
If I implement tonemapping in the shaders, wouldn't the alpha-blending be incorrect then since it would be done in a nonlinear color space?


Yes it's partially incorrect to do alpha blending in the tone mapped space : a semi transparent object drawn on top of an overbright object will not maintin the correct brightness, instead the shining object will look dull. But it's a compatibility trade off (something that allowed Half Life 2 to run on any shadermodel 2.0 card). This is true for any post processing effect (motion blur, depth of field, realistic bloom, haloing, fog etc.)

If you rely on high quality multisampling/alphablending you should do the Direct3D10 version, that is you should read the samples individually in a pixel shader after you've finished with the whole scene, tone map them and do the downsampling.

You should note that the problem you're seeing with edge antialiasing is also true for any filtering that happened before your tone mapping operator and that includes anisotropic filtering. An element that people often forget or dismiss totally. Anistropic filtering is designed to work optimally in linear space.

That problem with anisotropic cannot be fixed simply by doing the filter after the tone mapping (because it happens at texture read in a fixed function texture unit). Some level of supersampling might help.

---

There a third way, but I haven't seen implemented anywhere. I planned of doing a tutorial on it someday.. The idea is that you could maintain the true brightness ratio locally but not globally, that way things like anti aliasing that rely on the linearity but only on neighboring pixels could be applied at any stage of the rendering pipeline.

LeGreg

Share this post


Link to post
Share on other sites
Yann L    1802
Quote:
Original post by LeGreg
There a third way, but I haven't seen implemented anywhere. I planned of doing a tutorial on it someday.. The idea is that you could maintain the true brightness ratio locally but not globally, that way things like anti aliasing that rely on the linearity but only on neighboring pixels could be applied at any stage of the rendering pipeline.

Sounds interesting. Any more details ?

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  

  • Similar Content

    • By cebugdev
      hi all,

      i am trying to build an OpenGL 2D GUI system, (yeah yeah, i know i should not be re inventing the wheel, but this is for educational and some other purpose only),
      i have built GUI system before using 2D systems such as that of HTML/JS canvas, but in 2D system, i can directly match a mouse coordinates to the actual graphic coordinates with additional computation for screen size/ratio/scale ofcourse.
      now i want to port it to OpenGL, i know that to render a 2D object in OpenGL we specify coordiantes in Clip space or use the orthographic projection, now heres what i need help about.
      1. what is the right way of rendering the GUI? is it thru drawing in clip space or switching to ortho projection?
      2. from screen coordinates (top left is 0,0 nd bottom right is width height), how can i map the mouse coordinates to OpenGL 2D so that mouse events such as button click works? In consideration ofcourse to the current screen/size dimension.
      3. when let say if the screen size/dimension is different, how to handle this? in my previous javascript 2D engine using canvas, i just have my working coordinates and then just perform the bitblk or copying my working canvas to screen canvas and scale the mouse coordinates from there, in OpenGL how to work on a multiple screen sizes (more like an OpenGL ES question).
      lastly, if you guys know any books, resources, links or tutorials that handle or discuss this, i found one with marekknows opengl game engine website but its not free,
      Just let me know. Did not have any luck finding resource in google for writing our own OpenGL GUI framework.
      IF there are no any available online, just let me know, what things do i need to look into for OpenGL and i will study them one by one to make it work.
      thank you, and looking forward to positive replies.
    • By fllwr0491
      I have a few beginner questions about tesselation that I really have no clue.
      The opengl wiki doesn't seem to talk anything about the details.
       
      What is the relationship between TCS layout out and TES layout in?
      How does the tesselator know how control points are organized?
          e.g. If TES input requests triangles, but TCS can output N vertices.
             What happens in this case?
      In this article,
      http://www.informit.com/articles/article.aspx?p=2120983
      the isoline example TCS out=4, but TES in=isoline.
      And gl_TessCoord is only a single one.
      So which ones are the control points?
      How are tesselator building primitives?
    • By Orella
      I've been developing a 2D Engine using SFML + ImGui.
      Here you can see an image
      The editor is rendered using ImGui and the scene window is a sf::RenderTexture where I draw the GameObjects and then is converted to ImGui::Image to render it in the editor.
      Now I need to create a 3D Engine during this year in my Bachelor Degree but using SDL2 + ImGui and I want to recreate what I did with the 2D Engine. 
      I've managed to render the editor like I did in the 2D Engine using this example that comes with ImGui. 
      3D Editor preview
      But I don't know how to create an equivalent of sf::RenderTexture in SDL2, so I can draw the 3D scene there and convert it to ImGui::Image to show it in the editor.
      If you can provide code will be better. And if you want me to provide any specific code tell me.
      Thanks!
    • By Picpenguin
      Hi
      I'm new to learning OpenGL and still learning C. I'm using SDL2, glew, OpenGL 3.3, linmath and stb_image.
      I started following through learnopengl.com and got through it until I had to load models. The problem is, it uses Assimp for loading models. Assimp is C++ and uses things I don't want in my program (boost for example) and C support doesn't seem that good.
      Things like glVertexAttribPointer and shaders are still confusing to me, but I have to start somewhere right?
      I can't seem to find any good loading/rendering tutorials or source code that is simple to use and easy to understand.
      I have tried this for over a week by myself, searching for solutions but so far no luck. With tinyobjloader-c and project that uses it, FantasyGolfSimulator, I was able to actually load the model with plain color (always the same color no matter what I do) on screen and move it around, but cannot figure out how to use textures or use its multiple textures with it.
      I don't ask much: I just want to load models with textures in them, maybe have lights affect them (directional spotlight etc). Also, some models have multiple parts and multiple textures in them, how can I handle those?
      Are there solutions anywhere?
      Thank you for your time. Sorry if this is a bit confusing, English isn't my native language
    • By dpadam450
      FINALLY, upgrading my engine to openGL 4. I was having some trouble so I started with a stripped down application and was wondering if VAO's are required, because I have a sample working, but if I remove the VAO then it doesn't seem to like drawing my triangle.
  • Popular Now