Development Journal of a Raytracer

Ok, I tried to continue to model the character. I ran into considerable new problems, some already fixed. I tried to model the breasts (yes, the character is female) and didn't get anywhere. I managed to make so many errors, that I can basically throw away what I did the last 2 hours. But not all is lost, I at least now know what not to do.

I guess that is very valuable, as I know better now what to change for the next character. I guess only trying again and again and again will result in something good. But I will try again on this one first, the biggest problem was, that I tried to extrude multiple faces and then my program sometimes creates inner faces which are useless. I had quite some problems to remove those. The other problem is a defective modifier. The remove edge modifier always was wrong, I deviated from a principle to implement it. The reason was the usual, "it is simpler this way". As I had to learn, no, it isn't. It only results in bugs. So now I can just reimplement that part again and this time correctly.

This entry is not about raytracing, but web development. I have been reading a lot about it in the last weeks because I am still interested in it. There are many things I wonder about which seem to be largely irrelevant, at least considering i rarely stumble over other people thinking about it. The reason I am interested is connected to the character I am modelling. I for some time think about using the jMonkeyEngine to make it possible to upload some of my models. So that they can be viewed from different directions.

I have been working on the extrusions. The feature I missed in modelling the character was, to move vertices in the middle without leaving the extrude command. That way the steps would be connected quite closely and would stay editable. Well, I knew it was complicated to add that, but as I found out, the main problem is, that I don't understand the complete problem.

I haven't thought it through thoroughly and am unsure about quite some details. So I am constantly trying to find out, if what I am adding is indeed what I need or if I have done something stupid. The current version is working, but has a lot of problems. Most of them I added on purpose as I didn't know how to add it right. As I need some implementations, I choose the most simple one even if I am sure it is wrong. That way I at least now have a version I can try and see, if it is what I wanted. Then I can try to find out how to do it right.

But it also means I can't use it on the character model. I need that for some time and I don't want it to contain stuff from code which is as incomplete as the stuff I added last. As long as I don't store the stuff (or only store test cases I don't have a problem loosing) I don't have to think about converting the models I store now to a better format later. As I know, that the current data is wrong, I might have a very hard time to later convert it correctly, as I might very well use a bug as feature now and later find out, that that was very, very stupid.

As I really want the character to be finished as soon as possible, I will therefore continue on the model and perhaps only work on that new stuff from time to time a bit. As the character now doesn't depend on it, I can continue both in paralell without problems. So I will hopefully be able to use that new abilities of my program on a future character.

I finished the torso. At least for now. I had the basic form yesterday, but as I had feared I had created a ridge. Today I changed a rather complicated part of the program and edited the form the extrusion used to define the torso. Worked very well. But I need one other very complicated change in the program to continue. So my schedule is rather bad again, I fear.

Here are the images:

I am not entierly happy with the model, but I think it has to do for now. I finished the lower torso and indent to continue on the upper torso. I hope by the weekend I have at least the legs (complete with feet), torso and arms so I can start on the face. I would like to finish this model as soon as possible as I have some plans and I need the model first.

I managed to implement a mirror modifier. It was a bit harder and a bit easier than thought. There are some quite neat tricks I found to make it fast, but it also meant it is a bit more complicated than anticipated. But it is also much easier than I thought, because I found a very nice way to remove the no longer needed inner edges and faces.

The trick is, that the function throws away everything which is on the wrong side of the mirror plane, then mirrors the vertices and reconnects the edges and faces. So when I model I only have to make sure, that the "inner parts" are slightly at the other side of the mirror plane. And then they are just removed without a trace. Perfect for subdivisioning.

I would have posted an image, but I had problems to get the subdivision surfaces modifier to accept the object, it complained about illegal faces. I hadn't added code to handle big nonplanar faces correctly. Now it also can do that and I fixed an error for the big nonconvex faces as well. So I hope, I can render some shots of the character tomorrow.

And to be honest, there is a second reason I didn't quickly render an image, the model still has two major flaws and I first want to fix those before showing the rendering. I still suck at character modelling, although compared to my earlier tries it got a bit easier (and I think the model isn't totally awful). And this time I am not slow because I can't model, but because of bugs in my program.

But for the intended use the model should be enough.

I finally was able to model the lower torso of my character. No image as I am not satisfied with it just yet, some of the vertices have to be moved. Then I hope I will be able to post a picture.

I had to fight some problems to do it, though. There were again quite some little problems to solve and I had to add a new modifier to delete vertices. I had one to delete faces and one for edges. But up to now I never needed to delete a vertex. So implemented it.

Also I started on the mirror modifier to get a complete character. As is often done, I only model one half and then let the program mirror it and combine it to form one polyhedron. The shell of the modifier is created, now only have to add the parts that actually do the work.

I was a bit lazy today, but managed to fix lots of smaller things. One of them is, that moving in the draw plane modifier now works as expected. Also I fixed the "add faces modifier", it was not possible to flip the faces which were created. Now it is.

The other thing I shortly tried was to apply a simple cubic mapping to the temple and just mapped a terribly coarse bump map to it. It looked terrible. First because I created a terrible map, but also because there is a bug somewhere in the mapping code. So the test with a fast drawn mapping was actually quite enlightening. At least I now can look for another bug so it doesn't get boring. And I should now have everything to continue on the character. Finally.

I added some code so I can switch views without adding points. It wasn't easy, to be honest, but it works now. And I added a new operation to the draw plane to move around edges. It works, but not as good as I intended. I just copied the implementation of the other operations and so I only see a dot moving with the mouse and when releasing the mouse button the edges jump into place. I think I will fix that to move the edges with the mouse. That is how the other modifiers work and it makes modelling considerably easier.

So I hope, I finally have all parts together to model the lower torso of the character. I also thought about the UV-mapping for the temple and I think I will use a simple cubic-map for now. I hope that is enough and doesn't look too artificial. We'll see.

Fixed the small problem (missing normalization of direction on shadow rays) and was able to render an image. I think about mapping some bump maps to it (to give a slight impression of the walls being made of bricks) but I have a problem there, I have no good way to define the uv-mapping. I will see what I can do, but I actually want to continue adding stuff to finish the character. I will need proper uv-mapping for that as well, but not now.

I managed to extrude the walls and would post an image, but it looks horrible. The model is fine, the raytracer has some strange bug I have yet to find. I could reproduce it on a simple cube, so I hope to find out soon what is happening. It seems to be an interaction with the sky dome.

Also I found a weird issue with the triangulation, again. I don't know yet what it is. Might be simple (it looks like some triangles are simply misinterpreted as outside).

I fixed the two biggest problems with the DrawPlane and also found some minor ones. I didn't manage to model the walls of the temple, as I ran into an unhandled case. It was quite simple, I had snapped to a previous point and didn't notice it. First thing I did was to catch that case, so the program wouldn't crash, it would simply ignore it. But I will have to add a preview of the point which would be created when I clicked the mouse so I can at least see it. Then it wouldn't be a problem at all, I added the shift key as modifier to ignore snapping.

I had troubles to continue the character model. As I found out, my current tools won't be enough to work in a way, I am comfortable with. So I started to add a new tool I call "Draw Plane", it basically is a plane where I can set arbitrary vertices and connect them with edges. Sounds simple, but I needed several hours today to bring it to a state where it starts to work. There are still a lot of undefined operations (e.g. clicking onto an existing edge won't cut it into two halfs) and I also can't stop the editor (which is a bit funny) and so can't select the created edges to actually turn them into a face.

At least the last problem should be easy to fix. And then I will try it on a first model, not the character (there is still a little feature missing), but on the temple. I plan to draw some walls and then extrude them. I will use the new modifier for drawing the walls. I will also add it as a new constructor (so I then can really start an object from scratch). I am very happy with my progress, I did a lot of coding today, only counting the new modifier I wrote 1250 lines of code. And I fixed some bugs in other parts, wrote some support functions and so on.

So I hopefully will be able to render a new image of the temple tomorrow, with some walls, a bump mapping applied to them and the floor and I even have drawn a height map for a terrain to put it into. When playing with it I again noticed how bad my LoD-stuff on height maps is. I actually started to plan a new strategy for it and hope I will soon implement it. The biggest problem is, that it crashes quite easily and I honestly don't feel like going back into the existing code to fix it. Not only is it not working well, it is also a very complicated bit of my program.

The clip plane is working now. It was a bit more work than anticipated, but all in all it wasn't too hard. The biggest issues were, that I want elements which are created to manipuate the scene to still work. Examples are light sources or cameras. The second one was to get the element to correctly react to selections.

I worked a lot on the inner core of the raytracer. Three years ago (yes, nearly from the beginning) I decided, that it makes sense to integrate the transformations into the native raytracer representations. So if a cube was moved, the faces would be created at the final location. I did this to save on transformations, which are a bit expensive. But there is a problem, if you use the same object over and over (instancing), you need a lot of memory. Additionally it only works for primitive objects. Compound objects (like groups) don't work so well with it at all.