Anshuman Razdan wrote:
>
> Thanks for the pot shot at the academia. I am sure my collegues would
> appreciate that comment.
>
> Now to the point (without turning this into a flame war) - Yes, the
> prevasiveness of STL is not in question. The debate is whether to look ahead
> and cure the problems. Ofcourse STL format is so simple it can be translated
> into any new format that is designed. Even though bandwidth and memory is
> not that much an issue now (again we in academia are used to T1/3 lines and
> internet-2) the redundancy of vertices is still a waste and any good
> software architect will tell you that this kind is a drag on the performace.
> I make my academic case below:
>
> Lets look at some numbers:
>
> If there are N vertices, atmax there are 2N triangles in a mesh. For a STL
> file (binary format - minus the header of 80 bytes) its 50 bytes per face
>
> so total bytes come to 2*50 N bytes.
>
> For N = 1000,000 (1 million traingles)
>
> total bytes = 100 * 1000,000 = 100 x 1e6 = 100 MB of memory
>
> Now lets consider another format for comparison sake to generate the same
> information as the STL file (I am not recommending this format this is for
> comparison only). All the vertices are listed first so they appear only once
> in the file so there is 12 bytes/vertex information (keeping it the same
> i.e. floats). Its followed by faces. Each face has three unsigned int for
> vertex id (4bytes for the vertex id - a vertex id is generated from 0 to N
> where vertex appearing at 'l' would get id 'l') i.e. 12 bytes. No need for
> the normals since normals can be computed by doing a simple cross product.
>
> So per vertex 12 bytes - for N vertices = 12 N bytes
> Per face 12 bytes and for 2N faces = 12*2*N bytes = 24 N bytes
>
> In this case if 2N = 1e6 i.e. number of faces 2N = 1 million then number of
> vertices = N = 1e6/2 = 5 e5
>
> for 1 million triangles - i.e. 2N = 12 * 1e6 = 12 MB
> number of vertices 5* e5 * 12 = 6 e6 MB
>
> Total bytes = 12 MB (faces) + 6 MB (vertices) = 18 MB
>
> Hmmm that is less than 1/5th of the stl file. You dont have to do any
> checking whether two vertices that are off in 4th decimal place are really
> the same or not, etc etc etc. I wont even go into processing times and other
> software issues.
>
> This information provided as a free service of the academia.
>
> AR
When you carry this type of improved format one step further, it is easy to
maintain model topology. This enables very efficient slicing of a model
which saves time. Improved time to market isn't just an academic pursuit...
-- ======================================= Stephen J. Rock CII8015 Center for Automation Technologies Rensselaer Polytechnic Institute Troy, NY 12180 USA (518)276-8652 Fax -4897 http://www.rpi.edu/~rocks =======================================For more information about the rp-ml, see http://ltk.hut.fi/rp-ml/
This archive was generated by hypermail 2.1.2 : Tue Jun 05 2001 - 22:53:48 EEST