|

Since sIBLstage reached a point where its good enough for me, and commercial interest is low, I paused that project from the start of this year. For the end of this and the next I have 3 mayor programs I want to finish. During holiday, in a rushed manner, brought them all to a stage, where I:
a) see a promising future for them in my workflow b) can be finished by on/off coding on Saturday evenings. (The main concepts are mainly done, so most is meditative coding.
The PBR generator
>
Currently my PBR texture scanning relies on some unreliable software, or indie versions. Especially generating the height and normal maps. The generator already solves this, by estimating the light source….but I want to expand it further. Trying to get also the other maps out of this. The generation of a flat color map was a bonus I didn’t expect, but I have to see how well it reacts on some more complex shoots.
So its again a project which includes developing algorithm, and testing them with an input I have to capture in real life. (Similar to sIBLstage) which cost time, but it is fun.
The city generator

This one used AI for coding the heavy part (the L-System). Don’t get me wrong, my current stance of on this is for simple example projects and explanation for some processes its ok. But I just have limited time and energy in my free time, and learning and coding a L-System by myself would need at least two weeks of continuous coding, which currently I can’t invest. Sure AI coding makes you lazy, does not help to code better, and builds a dependency towards a group with commercial interest, which as a hobbyist is not a sustainable approach. But since I don’t see any monetary outcome of such a project, and I will never need such a process anyway for other projects, I have lesser problems using it. Anyway, currently I have around 180 parameters (and these will certainly will get much higher especially if I want to make the buildings more complex) and this does not include cross dependencies which has to be added, this will be a real long term project.
I see why a certain commercial solution out there relies heavily on scripting...which I want to avoid (although implementing a python parser in a project sometimes is on my todo list) The difference to other non commercial systems is, that I will put emphasis on European cities, and less on those skyscraper heavy US ones. So yeah, no immediate need, but in the past I wished several times I had something like this at the disposal.
The face match program

This one was on my todo list for a longer time. I always wondered if these face matching libraries can be exploited to mass produce humans heads. Most of them rely on a python environment which I want to avoid. But I found a decent c++ library and started the program.
It allows you to input several images of the face and matches landmarks / 3d geometry. From there it calculates a weight map (simple dot product to camera), which can be edited further by a brush. This weight map is taking into account when producing a weighted geometry and texture map from all shots. After that it either exports just the mesh, or modifies a more complex body wich features the same mesh geometry. In my initial test this manual weight brush is a must have. Its fun to paint out mismatched, or poor estimations in some parts, generate a new texture map and see the blend results.
This one went pretty far. And I might make a facet track application from the acquired knowledge in future. What makes this a long term project, are the features I still want to add:
FBX and Blend shape support: My plan is to allow you to select if an matched face contributes to the geometry, or just affect the texture or create a blend shape.
Unlighted texture: I want to implement some kind of HSV blending: if a texture value in one image is more in the 0.5 range than the others it should be taken more into account. My idea is to produce a more delighted version, if these shots are taken under different lighting conditions...maybe I need something more complex, but for now this is my plan. Maybe it needs a more complex steup, but we will see.
And last but not least add an additional dialog, which lets you export images from a movie clip and let you quickly assign them to one of the three states, or a combination of them. (EG use this grab for texturing, for blend shape or as an object contribution.
Well that’s it, that’s my 3 mayor projects I will work in the next year. So if I fall silent for multiple months, it might be because I’m working on those in my free time. And yes, this is if everything works out as intended, can't be sure in those uncertain times.
|