During my experience in various commercial and non-commercial projects I’ve came across some interesting CG ideas, algorithms and approaches. Some of them I get to implement on those projects, some of those are already implemented and I just have to use them. Those that are left behind are still lingering in my mind with the label “I wanna do this”. That’s how I’ve decided to create a playground for myself.
First, it would be used to have slightly more organized (than regular drafty piece of code while trying something out) collection of various methods, tools and algorithms relating to 3D rendering and CG. Although it is way slower than just to bash some lines of code, but trying to fit methods I want into somewhat nice framework and pipeline is part of the task. In other words, it’s not really a playground, but more like a tool for myself, were I can play around with new CG and 3D stuff.
Another reason why I wanted to do this is to have a playground for shaders. There are plenty of shader authoring tools on the vast Internet, but I still haven’t found one that would fit my needs. At the moment I am using a set of ShaderMaker, Shader Toy and Shdr. Having my own environment suited for my style and needs is something I was misssing.
I’ve started it as a very simple “load and show” program. Since I had OBJ file loader implemented some time ago (here), I’ve started with using/adapting/adjusting the same code. And, implemented simple Phong shading with forward rendering using OpenGL. I did some more stuff on top of that, such as some GUI implementation, some abstract classes to support extensibility to some sort, etc.
Finally, as my next step (and one of those “I wanna do it”), I’ve started with deferred rendering. And I had my ups and downs. Idea is extremely simply. But it took me a lot of time to figure out and find that rendering empty vector of light objects results in empty/dark scene. Stupid me. Anyways, picture above is current screenshot of my test scene rendered with 500+ point lights and couple of objects. This is still not optimized version of the algorithm so frametimes are just around 15ms.
Coming up next: the shadow! Probably. Well, first I will optimize the rendering and then most likely will want to go for shadows. Or some spotlight implementation and some other cool stuff related to that. Argh, so much stuff, so little time… 🙂
Alas, the first year of my master program has ended and I have decided to share one additional project I have been working on the last half of the year.
Actually, the course that I want to talk is Advanced Computer Graphics, which was extending through out both Study Periods of this semester, thus I am posting about it just now, when it is finished.
The course was essentially about doing a project work in a field of computer graphics with a chosen topic. And I chose spherical harmonics, because I was fascinated after our teacher introduced it. How powerful and fast it could be to make nice effects, such as world-space ambient occlusion and inter-reflection. Thus I have been working ever since the first seminars of this course in the end of February.
I’ve decided to try to work with OpenGL wrapper for Java – LWJGL. Thus I was able to work quite seamlessly both on my Windows and Linux installation (which I switched quite often). To begin with, I’ve wrote some helper classes to hold the information about vertices, vertex array objects and so on. But then the first problem occurred – I couldn’t find a decent model file loader for Java. Thus, I wrote one myself. I’ve spent quite a lot of time, but I wrote an OBJ file loader including with materials and textures. It is not fully finished (e.g. doesn’t support smoothing groups and such), but it was and is working for my needs. Later on I’ve implemented the classical rendering (the switch between classical and spherical harmonics could be seen in first part of the demo). I did that because I wanted to have some kind of point of origin, with what I could compare results of spherical harmonics (SH) rendering. And I wanted to test my OBJ loader without delving deep into SH calculations. After I’ve completed the classical rendering and OBJ loader, I’ve continued to work with implementation of SH.
I am not going to explain all the mathematics behind the SH because even for the course mates it was boring 😀 Thus I will tell what I have done with SH implementation. First of all, I’ve implemented all the methods required to calculate and use SH, but then, with increasing complexity of models, calculation times were going over the top. To construct SH for simple model with <1000 triangles would take about a minute or so. Thus I proceeded with research and implementation of ray tracing acceleration structures. That is construction of SH heavily relies on ray tracing from each of the model’s vertices. Because I’ve already had a small class representing axis aligned bounding box (AABB), I’ve decided to go for hierarchical AABB acceleration structure. According to my research it is not the most efficient structure, but it was efficient enough for me. After I have completed that, super complex models with hundreds of thousands triangles would be ray traced in a matter of minutes, thus it was sufficient enough. But still, the result of ray tracing is more or less pre-processing of the model.
That is why I have implemented SH saving/loading. Very simple and plain text format (I was inspired by OBJ format, thus very similar) which stores all the coefficients for SH for a certain OBJ file. If SH file is not found or is not representing the OBJ file (hash check! 🙂 ), all the SH harmonics are recalculated and saved again. This literally reduced pre-processing time almost to zero. Because of some complex models and high amount of spherical harmonics data, sometimes files with SH coefficients exceeds hundreds of megabytes, thus it takes a bit of time to load them up.
So, after core technology was done, I headed for the implementation of somewhat of a game. Downloaded several freeware models, created some myself. Created some what system for having complex models with separate parts, implemented interface, AI for enemy tank and even added fog effect so that background would blend seamlessly. I was almost ready to go. Almost.
After the supervision it was suggested for me to make something more, thus I have additionally implemented a skybox. It was rather easy to implement, but actually I was impressed about how it changed the feeling of the game. Because the skybox changed the ambient color of the background, I had to change the color of the fog and tweak the fog itself, but it was worth it. Then I have continued with the most interesting suggestion for me – screen-space post-process pass for volumetric shadows. Technically it is simple radial blur, but after some struggle with framebuffer objects I have implemented it and combined with the skybox it ended up to look super cool. At last, as a final touch I’ve added some textures to make it less plain and a small bar for displaying player’s health.
Below you can see the tech demo – a small video which I have compiled out of all small videos which I’ve taken through out the development.
Last week in Prototyping course we had a lecture and workshop on 3D printing. And we were planning to play some RPG on Sunday. And we were quite short on dice. That is how idea of modeling and printing my own custom die was born. The first thing I wanted to try is to make empty die with numbers on the sides and some cool thing (like small castle) inside. But then I realized that such kind of a die might be terrible unbalanced, that’s why I sticked just to the numbers on the sides. Here are some screenshots of modeling and the actual result. I am very very happy about this.
One of my first play around with 3D. Not like modeling or real 3D programing, just mechanics. Each time you start the game, it generates random maze with only one path to the far away standing “lollipop”. Good luck finding the way 🙂 (Controls are also WASD).