The 9/11 Forum

Wow.

How can we get renditions of WTC1 from the west? southwest? northwest?

That is the coolest tool. How can us little people use it?

Major_Tom wrote:How can we get renditions of WTC1 from the west? southwest? northwest?

Just place the camera where you want it to be, and make sure a light is shining on the target. With that last model, you can even look around the lobbies of the towers. I'm planning on recreating all the relevant camera positions of videos and images that have been, or will be, used for motion analysis. Later, I plan on recreating the collapses to whatever detail is possible.

That is the coolest tool. How can us little people use it?

I'm little people, too! I'm reminded of a line from Blade Runner:

"If you're not cop, you're little people"

and so it is.

True, some of these things are not so easy to set up and use. Some require programming your own application to use a library, like the NVidia stuff. Some of the NVidia stuff also requires a GPU and otherwise hefty graphics card. I bought my new workstation with this in mind - 8800 card and PhysX co-processor. But there's plenty of (free) stuff available that is ready to go out of the box, no C++ coding and no special hardware required.

A good full featured renderer is POV-Ray. The first rendering of this model I did, upstream a few posts, was done in either POV-Ray or POVMan, an offshoot of POV-Ray that supports Renderman type shaders. For other RIB-compliant renderers, there's 3Delight and Aqsis. One that can open the above model directly and handle the size is Kerkythea, which is a combo modeler and renderer. It may be the best bet for someone who doesn't care to write a scene description by hand in a text editor. Later, I'll put together links to all the relevant stuff.

I've been assembling as much physical metric data as possible - dimensions of buildings, placements and orientations, and so on to try to achieve correctness and realism. Even the highly detailed model of the plaza seems to have details of the WTC7 penthouse wrong. Corrections to huge files like this are difficult. Maybe not so much if you have 3DS Max but I'll have to rely on a programmatic parse to separate all these polygons into a hierarchy like building-component-location, then add/subtract polys and apply transforms to existing ones to get the right dimension. The models from Google's 3D warehouse are very nice, very helpful, and geo-located (not always correctly) but the actual dimensions used are off-the-cuff. Just load several versions of the towers simultaneously and you'll see the variation. In essence, everything will need some correction and tweaking.

Edit: the tool I used to render the translucent image is here: http://developer.nvidia.com/object/nvsg_home.html
There's a sample viewer that comes with the kit. It's NOT a full featured viewer but is good enough to do basic views. Its performance is far better than anything else I've seen. You can load the 3DS model after converting it to OBJ format, have to dig up links for that stuff.

Charles Beck's Jan 20 paper references my data, einsteen's too.

http://arxiv1.library.cornell.edu/PS_cache/arxiv/pdf/0806/0806.4792v4.pdf

Hi, Mr. JPA. You're attentive.

This post I made a while back was seemingly lame at the time. Off-topic, for sure, but not so lame after all. I was just looking at things the wrong way. This is the right way:

Statically indeterminate

On another thread, Major_Tom comments:

I didn't bother commenting on my observations because why would I waste my time if so few people would actually read it?

Bingo!

OneWhiteEye wrote:I want to draw all edges in a different color than the surfaces. Now, this is not the same as a cartoon shader, which applies a color to the silhouette edges, I need the edges formed by surface intersection to be outlines, as well as the silhouette. It would be like wireframe display but with opaque surfaces, backface culling, etc.

Have you tried using a texture with an outline and applying it 'per surface/triangle strip' ?


http://femr2.ucoz.com/photo/1-0-197-3

Model with a ceiling tile texture. UVW Map, applied with 'face' mapping. Instant space-shuttle

Using a texture with a picture of a box outline would be more what you are after I think, rather than a ceiling tile

Yeah, what I'm after is more like a drafting style outline where, if there's a real geometric edge there will be a line, not a line where the poly edges are. That's my problem, there are all these meaningless edges, artifact of the 3D construction process as opposed to representing something real.

I'm not as pressed to accomplish any of this as I once was.

OneWhiteEye wrote:Yeah, what I'm after is more like a drafting style outline where, if there's a real geometric edge there will be a line, not a line where the poly edges are. That's my problem, there are all these meaningless edges, artifact of the 3D construction process as opposed to representing something real.

I'm not as pressed to accomplish any of this as I once was.

It would be an almost impossible task for the model above, which is a mesh. For simple primitives the face map method works okay...


http://femr2.ucoz.com/photo/1-0-198-3

Hey, that's pretty nice! Very close, probably as close as can be. I may need to pick your brain on the particulars. You're doing this in OpenGL or D3D?

OneWhiteEye wrote:Hey, that's pretty nice! Very close, probably as close as can be. I may need to pick your brain on the particulars. You're doing this in OpenGL or D3D?

I use 3dsmax, but the principle should hold wherever...mapping the same texture to each face. It's just a texture of a square outline.

OK, I follow. I think. Just having the outline on a rectangle image as a face mapped texture puts the outline at the edge of the polygon, whatever the shape. More primitive CSG panels will undoubtedly be best, I think I got side-tracked by the nice 3D models, where I don't need anywhere near that level of detail. Not this sparse, but somewhere between this and wireframe-ish:



Not having 3DS, it can be a little overwhelming looking at the file which, as I recall, had to be converted to OBJ or something else just for me to examine it. I do have a 3DS lib, which I was going to use to identify the elements with vertical orientation, cluster by horizontal positions, and try to 'extract' the walls and/or obtain bounding boxes for prominent features....

Quite an absurd way to approach things to get a few boxes. Which is why I didn't do it.

Edit: let me explain. I have the foundation for what could be some nice routines for automating the process of discovering the camera locations and view angles settings based on edge matching. Now, a really clever person could bypass the image projection part altogether and do all the necessary vector calculations to determine the alignments. Given time and sufficient motivation, so could I, probably. Using routines that were already developed for other purposes seemed easier, though none of it seems easy enough now. Best bet is to twiddle manually until it looks right.

OneWhiteEye wrote:OK, I follow.

Aiii. That's the ticket.

Edit: let me explain. I have the foundation for what could be some nice routines for automating the process of discovering the camera locations and view angles settings based on edge matching. Now, a really clever person could bypass the image projection part altogether and do all the necessary vector calculations to determine the alignments. Given time and sufficient motivation, so could I, probably. Using routines that were already developed for other purposes seemed easier, though none of it seems easy enough now. Best bet is to twiddle manually until it looks right.

If you want me to strip the models down to the bare minimum in 3ds, no probs, just let me know...what buildings, and how primitive.

Have been doing a lot of camera matching for the 175 orientation paper...first half draft just uploaded for a bit of feedback on t'other thread. The hardest part I rekn is dealing with all the foreshortening and depth of field wierdness that camera lenses cause...

femr2 wrote:If you want me to strip the models down to the bare minimum in 3ds, no probs, just let me know...what buildings, and how primitive.

Maybe so. Thank you for the offer. There was a time when it would have been a resounding YES, but I think the sensible thing to is to make sure I have a plan and intention for the results, so the effort doesn't go down a hole.

Have been doing a lot of camera matching for the 175 orientation paper...first half draft just uploaded for a bit of feedback on t'other thread. The hardest part I rekn is dealing with all the foreshortening and depth of field wierdness that camera lenses cause...

See, that's one of the reasons I was thinking about automating. The degrees of freedom involved make visual adjustment go pretty slowly. Throw in a little zoom or pan, and it gets even worse. Still, manual is probably the way to go.

OneWhiteEye wrote:See, that's one of the reasons I was thinking about automating. The degrees of freedom involved make visual adjustment go pretty slowly. Throw in a little zoom or pan, and it gets even worse. Still, manual is probably the way to go.

A very daunting task to consider automating indeed. Have to say I think it may be an impractical challenge, unless the end result was going to hit the market with a large price tag and a lot of copy protection

Have you ever used SynthEyes ? It will perform three dimensional mapping from source video and spits out scripts for many modelling packages. Tracks points then 'solves' for 3D. Allows for movie compositing between real footage and inserted models...bit of a learning curve though...

It still struggles/fails badly without some manual intervention for most of the challenges I've set it. My archaic version of 3ds won't take the scripts, but perhaps it may be useful for you...

http://www.ssontech.com/synsumm.htm

femr2 wrote:Have to say I think it may be an impractical challenge, unless the end result was going to hit the market with a large price tag and a lot of copy protection

Yes, I think you're right about that. Never was very good at estimating amount of work!

Nevertheless, there is some advantage to doing manual alignment with edges enhanced and faces de-emphasized, video and 3D scene, it's easier on the eyes.***

Have you ever used SynthEyes ? It will perform three dimensional mapping from source video and spits out scripts for many modelling packages. Tracks points then 'solves' for 3D. Allows for movie compositing between real footage and inserted models...bit of a learning curve though...

It still struggles/fails badly without some manual intervention for most of the challenges I've set it. My archaic version of 3ds won't take the scripts, but perhaps it may be useful for you...

http://www.ssontech.com/synsumm.htm

Wow, the cool stuff that's out there these days. Yes, I'll have to look closer at this. Seems they go about it in the right way, not the BS backdoor way I was considering. It is a reasonable price.

***Edit - this is a prime example, an edge-filtered video against a simple trapezoidal prism as 3D figure would make the alignment easier, no matter if human or machine as operator:

http://the911forum.freeforums.org/technical-notes-on-video-motion-analysis-t55-270.html#p2306