EDITORIAL NOTE:
This Draft Document is in the process of being written to support the upcoming release of MakeHuman©.
An earlier version of the document can be found at http://makewiki.aleppax.it/pmwiki/pmwiki.php/User/TheHumanoid3DModel .
All of the graphics on this page will need to be updated before this document is released.

Introduction

The principal aim of the MakeHuman project is to develop an Open Source application capable of realistically modelling a very wide variety of human anatomical forms in the full range of natural human poses from a single, universal mesh.
Central to this is the design of a 3D humanoid mesh that can readily be parametrically manipulated and deformed to represent alternative anatomical characteristics while retaining and respecting a common structural skeleton that permits poses and the corresponding deformations to also be parametrically manipulated. This objective has been pursued to afford the artist the maximum degree of experimental freedom when using the software. It frees the artist from the artificial constraints that are inherent to a model that has pre-established gender or age.

By pursuing this aim the MakeHuman Team have developed a model that can combine different anatomical parameters to transition smoothly from the infant to the elderly, from man to woman and from fat to slim. The vast wealth of potential combinations provides the artist with an extraordinarily broad range of possibilities for artistic expression but presents many interesting problems to the development team. In particular it adds to the classical problems of 3D modelling (number of polygons, square or triangular faces, etc.) the problems of constructing a super mesh that can be transformed into any form of human while being sufficiently optimised to be able to be manipulated on desktop machines, yet still producing a professional quality of output. By contrast, most modellers only have to produce a single model that can be created and adapted for a single project.

The current MakeHuman mesh has evolved through successive iterations of the MakeHuman project, incorporating lessons learned, community feedback and the results of considerable amounts of study and experimentation. No generic mesh is perfect and this mesh has inevitably been subject to some compromise and will undoubtedly continue to be refined in future releases. Nevertheless, the current mesh represents a remarkable achievement and is a great source of pride for the MakeHuman team. The current iteration, known as the 'HM04' comprises a state of the art universal humanoid model. This paper describes the characteristics and capabilities of the mesh along with a brief history and discussions about potential future enhancements.

Evolution Towards a Universal Model

The evolution of the mesh through successive iterations illustrates a number of interesting concepts that have been explored and the understanding that has been encapsulated into the current mesh.

1. The first, very interesting model, HM01, was realized by Enrico Valenza in 2002.
2. The second remarkable mesh (K-Mesh or HM02) was modelled by Kaushik Pal in 2003
3. The third mesh was modelled by Manuel Bastioni upon the HM03;
   
4. The fourth mesh was modelled by Gianluca Miragoli (aka Yashugan) in 2007 and builds upon the experience gained on the preceding versions (Y-Mesh or HM04)
   
5. The fifth mesh build upon the previous one by Gianluca Miragoli and Manuel Bastioni (HM05)
6. The sixth mesh build upon the previous one by Gianluca Miragoli.

The knowledge gained over the years has driven the simplification and optimization of the model. A highly sophisticated and detailed model can be good for static, one-off models, but an application designed for real-time manipulation of related groups of parameters and for real-time visualisation needs to be efficient. The MakeHuman solution to these contradictory pressures is to create a simplified, optimized model and to support the generation of sub-surfaces that can be used to smooth out imperfections before rendering.


Fig. Head history

Fig. Comparison between K mesh and previous model.

Conception to HM04

Todo...

Starting Posture

One of the subjects that generated considerable discussion during transition between HM01 and HM02 was the starting posture for the mesh. Alternatives ranged from the classic outstretched 'Vitruvio' posture to a near fetal position that was advocated by some artists.

The errors introduced by the deformation of a mesh when bent or twisted away from the original position become more evident the further the mesh needs to be deformed, so selecting a starting-point where each joint is partially bent could reduce the amount by which the mesh has to be deformed to reach the maximum and minimum limits of that joint. However, this would provide a bent and buckled starting posture nearly in the fetal position that is far from a natural position for anyone to adopt under most normal circumstances and which is not a particularly good position for artists to start from when attempting to model a real-world pose.

As illustrated in the figura above, selecting a midpoint for the starting posture can considerably reduce the most extreme amounts by which a mesh needs to be deformed and can therefore reduce the worst of the resulting errors.

However, experience gained from the original mesh revealed that this choice is not as advantageous as was originally thought. While it can prove useful for the simplest articulations, like the elbow and the knee (provided that it takes account of the contraction of the outer surfaces by sufficiently spacing out the vertices), in more complex joints with multiple degrees of freedom, like the shoulder joint, the results can even be disadvantageous:

• Working with a joint with 3 degrees of freedom and a starting position that is partially bent can make it very difficult for an artist to visualise the joint rotations required to achieve a correctly modelled result
• It can be difficult to control the way the mesh buckles around the outside of the joint
• It makes it difficult to develop the model for the surfaces folded into the joint (e.g. under the armpit)
• The detection of topological errors during mesh development becomes more arduous It's much more inconvenient not only to work with the shoulder already lowered, but it's also extremely difficult for the mesh to buckle well for the opposite positions to that of departure, for example that of an upward extended arm. Moreover it was difficult to model the areas under the armpit, and also the perception of topological errors becomes more arduous.

Details

The K-Mesh introduced several detailed elements which, even though they are realized with a very low number of polygons, add considerably to the realism of rendered characters. For example, the inside of the mouth has been precisely modelled to provide an anatomical correct structure that can be adapted to represent a very wide variety of human mouths. This includes teeth, gums and a tongue.

A Universal Model

The next challenge was to construct a universal topology that retained all of these capabilities but added the ability to interactively, programmatically adjust the mesh to accommodate the variety of anatomical varients found in the human population. This challenge could have been addressed by dramatically increasing the number of vertices used for the mesh, but the resultant, dense mesh would have limited the performance on all but top end machines and, even with extremely powerful computers it is generally recognised that an optimised mesh is preferable to one containing useless or morphologically insignificant points because:

• A more economic number of control points supports more orderly and precise modelling, avoiding the confusion of edges inherent to a more dense model
• The savings on processor and memory resource can be better invested in providing more sophisticated functionality and greater fluidity to the artist
• The lighter model better supports the possibility of incorporating larger numbers of characters into a rendered scene

Mirroring the discussions on the initial pose there were discussions on the ideal base anatomy that could be transformed into the various extremes of the human anatomy. These discussions resulted in agreement that the initial mesh should occupy a middle ground, being neither pronounced masculine, nor pronounced feminine, neither young nor old and having a medium muscular definition. An androgynous form. A form midway between male and female, old and young, thin and fat, muscular and lean.

Mesh Data

Vertices	11787
Texture coordinates	12223
Faces	12284
Groups	347
Quads	10588
Triangles	1696