Essential UV on subdivide Surface

With the increasing use of CCSS Catmull-Clark subdivide surface (CCSS), problems have emerged. However, in general, it is better to use CCSS than simply using polyhedron or OpenGL. In the first place, there is no problem with the combination, deformation, skeleton binding, and Paster of the OpenGL, after all, my personal productivity is no longer comparable to that of bluesky's entire R & D team, where there are some Ivy League graduates of pH. d. Polyhedron has obvious advantages. It is easy to shape, easy to learn, and has no mathematical requirements. The most disgusting problem is its inefficient Performance of High-Precision models and its terrible memory usage, it only suits Mental Ray's weird appetite. The advantage of CCSS is that everything is done in on-fly, and renderman 12.0 has improved CCSSAlgorithmTo make the use of CCSS popular.

Currently, the wide application of zbrush and mudbox has enabled most studios to use traditional modeling tools such as 3 DSMAX and Maya to generate low-precision control polygon, then, you can directly access the zbrush or mudbox to enrich the model details and compute and generate the displacement texture. However, this process involves many basic computational geometric issues, that is, how to ensure the correctness of ing. From a mathematical point of view, from 2D UV space to 3D XYZ space represents such a relationship,

• And

It can be explained in vernacular, that is,

Surface is a function in space R ^ 3.

UV is a function in R ^ 2 space.

There is a one-to-one ing relationship from UV to surface, and the reverse ing is established, that is, they are all functions and mutual inverse functions.

 

Article 3 is the origin and manifestation of all model-related problems. "A general surface can not be parameterized without distortion." As said before, not all surfaces can be parameterized without deformation. In the actual CG production process, the modelers do not create a general surface, so they also choose to convert model parameters into their preferred methods. For example, for the head model, cracks have always occurred frequently. Because we can't always ask the artist to use the basic knowledge in topology to think of the human head as a balloon. When we divide UV, we can think of it as a tear, but not a tear, place the grid evenly in the UV area. It is too difficult, not intuitive, and difficult to draw textures. They prefer to face a face, a neck and a neck, and a UV distribution of the ears, rather than a distorted face covering the whole plane. The methods proposed in Hugues technical PE's spherical Parametrization and remeshing are commendable, but they are unlikely to be popularized in actual CG production. It reminds me that most of PBR (MC) is used for research-based rendering, but not many people in the vfx industry will use it.

Regardless of the concept, artists and modelers Need To Know That filtering is never the best solution for matching the signal frequency between the UV space and the model space. There is a fatal problem when using CCSS, that is, what you see is not what you get (was Du hast gesehen ist nicht was Du willst erhalten). You are confident before rendering, after rendering, the UV does not match, the UV does not match, and there are cracks. It's really terrible. It's useless to ask the original military doctor on the pole. interpolation can be improved, but it can only be congested. At the same time, we often see this situation, the model is complete, UV textures are torn, such as the head, the most complex is the hand. Such a model has a slit no matter how it is, and how to adjust the texture or interpolation is just a temporary solution. The slit is still there, but it is not so obvious. Because the boundary of the model is not necessarily the boundary of the UV space. It may be solved through adaptive boundary sampling, but it seems that this advanced function is not available for commercial software that is widely used at present. According to my discussion with Professor Peter Schröder of Caltech, we cannot directly use subdivision algorithms to deal with UV limitations here, because UV space and model space are not one-to-one, generally, the number of Uvs is more than the number of vertices.

But in fact, Maya has comprehensive internal support for CCSS, but it has not been explored and used. Of course, for traditional modeling tools like 3 DSMAX and Maya, the temptation of new tools is unstoppable, but it may not be one or two pH to integrate with the existing production process and properly solve the problem. d.

I will provide a pilot command lineProgramThe implementation is similar to the uvsnap function in Maya, but I generate a comparison chart between the UV range of CCSS and the UV range of the original control polygon, which is provided to the artist for reference, in this way, the Paster will more closely fit the CCSS surface generated by renderman to reduce the error. This program needs to be executed on a GPU that supports OpenGL 2.0, because the gpu I use is raster and edges directly. You can select the size and subdivision level of the generated image. This is a sample image. This tool has been used stably in our process. What's wrong with the command line program? Have you ever seen that renderman is a command line program? Hum ...... This is a sample image. The blue is subdivide UV, and the red is the UV distribution of the original control polygon in Maya.

Please do not have any associations ~

It is used for internal models of maxcompute studio and provides reference to artists ~