I have update the design of this website. It should now be cleaner and easier to read. Also it is now using html5 and some advanced css features such as @font-face. It works nicely in firefox, google chrome and even on my nexus one (android phone), didn't even test in IE. I plan to make some more changes over time. Adding a page that lists all my experiments and a gallery for the graphics I did will probably be the next step. I'm also working on a new javascript demo. Feedback is welcome.

By the way, I think this is the first time ever my website has a white background.

Last Saturday I have released my little javascript pathtracer that rendered a very simple scene. This time I want to show off some of the more advanced effects that can be simulated nicely using path tracing. I decided to set up a scene similar to the well known Cornell Box to demonstrate effects such as soft shadows, color bleeding, reflection, refraction and caustics. I did not include Depth of Field in this experiment because it didn't fit the scene well in my opinion.

The experiment

Click the image to start the rendering. Warning, this experiment is slow. Run it using a fast web browser (google chrome) and be a little bit patient. Opera won't work because it doesn't support web workers.

You might also want to check out my previous path tracing experiment.

Path tracing

Path tracing is a way of solving the rendering equation using monte carlo integration. It is a form of ray tracing. According to wikipedia 'Path tracing is the simplest, most physically-accurate and slowest rendering method'. Sounds like the perfect target for a javascript experiment!

Implementation

The implementation is basically an extended version of the path tracer in my last post. It is using html5 web workers to render the image on up to 4 cores without locking up the page. One of the cool things I came up with is jittering the points on the view plane for each pixel to get rid of aliasing. I also changed the color of the light to that of warm sunlight (~5400k). I probably got the fresnel equations wrong and generally things are quite rough. I'm not an expert in the field of raytracing and I probably did a lot of things in a suboptimal way. But feel free to have a look at the source code (~300 LOC) and ask questions. I think the code should be quite easy to understand.

Please write a comment if you've got anything to say. If you think my experiments are interesting, follow me on twitter or subscribe to my atom feed.

Improved version

An improved version of this path tracer is available here.

The experiment

Click to start. If it doesn't work in your browser and it's not IE or Opera, please write a comment. The output of the javascript error console would be very helpful too.

Path tracing

Path tracing is a way of solving the rendering equation using monte carlo integration. It is a form of ray tracing. According to wikipedia 'Path tracing is the simplest, most physically-accurate and slowest rendering method'. Sounds like the perfect target for a javascript experiment!

Implementation

My implementation is currently only calculating the diffuse term of the rendering equation. The only light source is the sky. The big sky speeds up the rendering and convergence quite a lot.

The sourcecode is split up into two files. main.js is the glue for initializing the workers and aggregating the results. worker.js is where you'll find the real meat. I'm spawning four workers right now to take advantage of modern multicore cpus. Feel free to ask questions if something is unclear.

Performance

As you can tell from the sourcecode it is clearly not optimized for performance. What is still interesting though is how various webbrowsers perform. On my Intel Core i7 machine google chrome (4.1) is roughly 26 times faster than firefox (3.6)! Chrome does 13 iterations/s, Firefox does 0.5. Also firefox seems to support only 3 workers, this leaves one core almost idle. The dev version of chrome is even faster at 18.75 iterations/s. My congratulations to the chrome and v8 developers. I hope firefox will catch up soon.

Twitter

On a side note, I created a twitter account for those who prefer it over the rss/atom feeds.

Update

I found a little bug in the code that affected the performance. I'm now getting more than 20.5 iterations/s in chrome dev. I think a faster random function would make it even faster.