He is my technical rebuttal. Kirby, I'll be happy to debate you point by technical point. So far, you haven't refuted one of my technical points-- just a lot of babble because you have a vested interest. I have no vested interest other than wanting to put a stop to Synthonics' propoganda machine, as it is bad for the photogrammetric industry.
I didn't realize I could post directly to the thread, so I will post my private messages to Bob and Oberjuran:
Bob,
I'll copy my email to Oberjuran to you below. It is up to you if you want to ignore my technical expertise. I was just trying to warn you all but it seems you are easily fooled.
As for 2D morphing (Kai's etc.), that is different than 3D morphing. It is irrelevant relative to Synthonic's patent though because they do not claim any patent on 3D morphing. I studied their the 8 claims of their recently awarded patent, "Method and Apparatus for Creation of 3D Wireframes". They like to call it a 3D morphin patent because they are repositioning the vertices one-by-one of an existing 3D geometry (mesh). But, this more akin to tediously building the vertices from scratch. The only step that is saved with their method is the marking of the polygons, since the mesh topology (polyongs) are already defined. Thus, user drags the vertices in 2 photos instead of marking them in two photos. Very trivial difference and besides this dragging of repositioning of vertices using photogrammetry was already published in 1991. I don't want to mention the research paper here because I'll keep it secret in case Synthonics ever sues me for infringment. By the way, the identity of my company will become publicly clear to all of you once we release our 3D human face product later this year. But, because of the nature of these statements, I prefer to remain anonymous as "Henry". Some of us have gotten very tired of Synthonics' propoganda machine and have decided to use this forum as a place to set the record straight.
Also, let me say for the record that Colin is not my friend nor have I ever known him. I simply sent him a private message and he was kind enough to post it. Please blame me for all my comments, not him.
=======================
Oberjuran,
Please post this message publicly on SI for all to read. I spent about 1 hour to write this so please let my effort be shared with all.
I only have a few days left on my trial SI membership, so if there are any more questions for me, please make them soon or email me anytime in the future at segovia@weblinq.com.
Since my message was posted publicly by Colin (thanks Colin), I acquired copies of the Synthonics' patents and have studied 2 of them in detail.
To answer your points quickly,
1. About Smithsonian, I don't have any information on that. But, I bet if you dig you will find that Synthonics' history of telling half of the truth will come out again.
2. Photogrammetry (the very old science of Synthonics' technology) is slow compared to laser scanning. True, it can be done on a PC. The problem is that every 3D point in the geometry of the objects has to be marked on at least 2 photos. Since even crappy (low resolution) VRML objects have around 300 points, you are talking about several hours for an object that doesn't even look that realistic. For a box shaped object, photogrammetry is reasonably efficient because only have to mark the corners. However, since most interesting objects have some rounded edges or rounded surfaces, usually 500 - 1000 points are needed to get something descent that could be used in a game. For movie production, anything under 5,000 points probably won't be accepted by the viewer. Take a look at Viewpoint's models which are used in movies such as Jurrasic Park and Godzilla (www.viewpoint.com) and you will see typically they have 20,000 or more points. You could not do these models with photogrammetry. It is just too many points to mark and keep tract of.
Face it, if one has a $100K contract (not to mention a $2.5 million one), then one can very well afford to spend $20K on a laser scanner which can scan about 10,000 points per second and an entire rounded object in 17 seconds. The money saved in labor will pay for the scanner in short time.
If you would like more info on 3D scanners (digitizers), see the June 1997 issue of Digital Magic magazine article entitled "3D Digitizers". The devices reviewed there are:
* Cyberware, $15K, www.cyberware.com (this one captures photo texture also)
* Faro, $5K, www.faro.com
* Immersion, $3K, www.immerse.com
* Polhemus, $6.5K, www.polhemus.com
* Model Maker, $40K, www.3dt.com (this one captures photo texture also)
* Digibotics, $55K, www.digibotics.com
Since that article is old, it doesn't mention the newer devices such as the $20K MetaStream Real 3D scanner manufactured by Lockhead (the aircraft company), www.metastream.com. The MetaStream scanner integrates well with their new MetaStream format for the Internet that allows real objects to be viewed in full 3D photo realism. This is a joint development with Intel and Microsoft. Synthonics has no chance against this technology alliance.
********************
Now to go into more detail on the Synthonics patents. First of all, the reason I went into such effort is because my software company is producing a software product (which we will release this year) that makes 3D human faces from a photographs. This product will be widely marketed and will likely blow away any chance for Synthonics to profit on the "3D morphing" patent. Thus, I had to evaluate the Synthonics patent to make sure we would not infringe. It turns out that the Synthonics "3D morphing" patent (Method and Apparatus for Creation of 3D Wireframes) has two severe weaknesses:
1. The unique part of the invention is trivial. It says basically, "instead of marking a point on 2 photos, drag the point instead so that we can already know the mesh topology"
2. It violates at least 3 instances of "prior art" that I found. "Prior art" is a legal term in patent law meaning previously published work.
So, not only would we not use Synthonics' method because it is so trivial and weak, but even if we did, Synthonics' patent would be easily overturned in court since there is prior art. The US Patent office does not have the responsibility of checking all prior art on a patent application. This is left to the defendents and courts if a case is ever filed.
All Synthonics' patents are based on the "3-pt orientation" patent. That is the meat of their technology. In fact they erroneously claim on their web site that their 3-pt method is unique and more accurate than other "6 pt methods". My photogrammetric expert contacts and I have looked closely at this patent and have found that although may stand up in court, it is a very weak method. It is basically just a complication of Church's 1950 resection method.
The truth is that Synthonic's 3-pt method is very inaccurate and won't work with a variety of cameras lens. So, the base of Synthonics' technology (the orientation of the cameras from photos) is weak when compared to the technology of other photogrammetric firms. Synthonics is not the only company with photogrammetric technology that can be used on a PC. See also 3D Construction (www.3dcoonstruction.com) and Eos (www.photomodeler.com). In fact, 3D Construction's 3D Builder product can also do 3-pt orientation in addition to many other types such 4-pt, and parallel lines. Eos PhotoModeler is perhaps the most flexible technology as it can do 3-pt (as control points) and it can also do 6 ***unknown*** points. In other words, with Eos's technology you can start with photos where you didn't even measure the distances between any points. With Synthonics', the distances between 3 points had to be measure before the photos were taken.
Synthonics' uses their patented 3-pt orientation as the final orientation of the cameras in determining 3D coordinates. In other words, since their is no statistical global optimization calculation, as is done with Eos and 3D Construction, the accuracy of Sythonics technology is usually only about 1 in 100, versus anywhere from 1 in 500 to 1 in 20,000 for Eos and 3D Construction.
One might argue that 1 in 100 is good enough for VRML models, yet it probably isn't good enough for museum archiving and movies. And Synthonics' technology won't work with a 35mm camera (the accuracy would be unreasonably poor). It needs 50mm or greater because it assumes the camera is a pinhole ideal. So, just read Synthonics' lie on their web site about how their technology is more accurate than statistical methods and you we have your first example of their propoganda.
I could go on and on...
Henry
segovia@weblinq.com |
