Implementation of SSAO for Computer Graphics & 3D exam
This project was developed as an assignment for the Computer Graphics & 3D exam, CdL Magistrale Ingegneria Informatica at University of Florence.
The project consists in an implementation of Screen Space Ambient Occlusion (SSAO)
as first described by Crytek and implemented in their famous PC videogame
Crysis, from 2007. The actual implementation
was never released but a high level description of the algorithm can be found in [1], in Section 8.5.4.3 (at the time of writing, a valid link to the paper is the following: Finding Next Gen – CryEngine 2).
Ambient Occlusion is a global methodology employed to compute to which degree a certain part of a 3D scene is affected by ambient lighting. In a basic model, every object in the scene is equally exposed to ambient light, but this is not realistic as many regions are in fact occluded: geometries surrounded by many objects, corners, tight gaps between objects, creases, etc. These regions are traps for scattered ray lights as they cannot easily escape: as a consequence, these areas appear darker than unoccluded geometries.
The objective of Ambient Occlusion is thus to reproduce this shading of the ambient color in occluded regions, in order to give more realism to the scene and add a natural feeling of depth.
SSAO aims at approximating the effects of real Ambient Occlusion, obtained with global illumination, in a real time environment, where the exact calculation would be infeasible.
SSAO is done on a screen space basis (hence the name) using post processing and deferred rendering. The steps involved are roughly the following:
- render geometry
- for each fragment in the geometry render target:
- reconstruct the view space position of the fragment
- sample some points in view space in an hemispheric volume oriented towards the surface normal at that fragment
- for each sampled point, project the point in clip space and test whether the projected point is behind the actual geometry in that clip position:
- if the sample is in front of the geometry, the geometry does not occlude the original fragment
- if the sample is behind the geometry, the sample occludes the original fragment
- average the contribution for each sample to obtain a single value per fragment
With this algorithm we obtain a screen space map of values in [0,1], where 1 means no occlusion and 0 means maximum occlusion. These values are then multiplied with the rendered geometry buffer, so as to attenuate the ambient color intensity by a factor proportional to the ambient occlusion map. Thus, the occluded regions are darkened, as required.
This implementation is realized in Javascript with the aid of the 3D graphics library three.js, which uses WebGL as rendering API.
The project requires Node.js and can be installed (together with dependencies) with the following commands:
git clone [email protected]:ivan94fi/cg3d_project_ssao.git
cd cg3d_project_ssao
npm installAfter installation, the development server can be executed with
npm run devand the production server (using Express.js) with
npm run build
npm run startAn online version of the project, hosted on Heroku, is accessible at this link. Please be patient on startup because the Heroku machine takes a while to fire up.
The application code follows the structure of the SSAO example in three.js, while the realization of the SSAO shader is based on the material available in the following sources, which have been of great help:
- John Chapman:
- LearnOpenGL:
- Matt Pettineo:
- OGLdev:
- Inigo Quilez:
[1] Martin Mittring. 2007. Finding next gen: CryEngine 2. In ACM SIGGRAPH 2007 courses (SIGGRAPH ’07). Association for Computing Machinery, New York, NY, USA, 97–121. DOI:https://doi.org/10.1145/1281500.1281671