Some Ideas of the company

It is really to fill in the feeling of emptiness and boredom, and pick up the things that have been studied before. Every time I think that Looking back on the past, I will have a new experience (not limited to some of the skills mentioned here). Maybe this is a good thing, right?

In terms of silhouette detection, we previously felt that brute force search was enough. In fact, based on the consistency between frames, we added some features, such as the angle between Edges smaller than a certain threshold value, candidates can be filtered out from a large number of triangles and searched as seeds, which is much more efficient ).

Besides, when detecting, we should follow the searched silhouette edge as the clue and search in the adjacent side or adjacent area to generate a continuous path, it can be either the screen coordinate system or the world coordinate system. When I used to sum up this experience, I always felt that deep-priority search would require recursion or something. Now I think it was complicated. In the end, I don't need to use traceback to work on the stack ...... In short, the generation of this path is inseparable from the subsequent parameterization.

The definition of silhouette edge can also be more accurate. The previous definition is that the value of N dot V of two adjacent sides of the edge is the opposite, that is, silhouette edge. But in fact, the true silhoutte edge should be all N dot v = 0. The polygon approximation model can only approach the original model even if there are more faces. Therefore, when n dot v = 0, no edge is found ). So there is a concept of interpolation. The method is to record the N dot V of three vertices on a plane instead of using the surface normal as the data source, but using the vertex normal as the data source. If the symbols are not all the same, then there must be a silhoutte edge passing through the two sides of the triangle and reaching the adjacent triangle. In this way, the silhoutte edge becomes the silhoutte contour, which is a segment defined on the triangle surface. This segment can eventually be connected into a ring (! ). For details about the two endpoints of this segment, interpolation is performed based on the result of the three n dot v s and the vertex position of the triangle.

The visibility of the Silhouette path is also a big problem. I think everyone is using ID reference buffer to render mesh and silhouette into this buffer, so that zbuffer can solve the visibility. After that, you need readback (the biggest bottleneck), and then compare the effective silhouette, raster, and raster with the color in the ID reference buffer to obtain the actual visible part.

Another point is to talk about crease edge, which was not paid attention to in the past for various reasons. In fact, it can be pre-calculated (view independent), as long as the angle between the two sides is greater than a certain threshold, it can be recorded as crease edge. Crease edge is very important in expressing the details of objects, especially if you want to use dotted lines to represent the details of invisible objects, without the crease edge, it will be very embarrassing (jot's tutorial 2 ).

Above.