How can I make my own Maltese Falcon?

Factual Questions
1

You’ve seen the movie, right? Haven’t you said to yourself, “I’d like to have my own Maltese Falcon”? Of course you have.

So you went to eBay, and found these cheap knock-offs that have apparently been copied not from the 1941 film’s prop, but from a completely different one made for the 1977 parody film, The Black Bird. IMO they look like crap, and completely lack the character of the real thing. Oddly, these copies are more detailed than the original film’s prop, which seems worn down by comparison.

So, since only the best will do, it would seem that we’ll have to make our own. (The owner of the site in the last link has already done this, but only made a dozen, all of which are gone. Yes, I’m planning on asking him how he did it. But in the meantime, I hoping you all have some good ideas.)

Note from that last link that the film includes a sequence that gives a near-360-degree view of the dingus, and several more shots that give good closeup views.

So does anyone know if it would be possible to take frames from the film, scan them into a computer, create a virtual 3D model, and then use that to create a real 3D item from which a casting mold could be made?

(For those who are concerned about the legality of this, let’s deal with that in a separate thread.)

Is it possible? Would it be incredibly expensive? Can I do parts of it myself? All of it?

Thanks.

Mods: although this post deals with a film, I’ve placed it in GQ instead of CS because I need the knowledge and experience of the scientists, technicians, geeks, and nerds who hang out here. Those artsy-fartsy types over in CS probably wouldn’t have a clue. :smiley:

2

Here’s the thread on the legal questions.

3

First, go to Malta. Then find a falcon and get her drunk…

Okay, now to read the OP…

4

Actually, the best way is to get a lump of clay and sculpt one yourself.

5

Summer break is coming in the US. Contact a local art teacher at the high school level and pay him (or some student he recommends) to make the thing for you.

6

Co-incidentally, my research at the moment is doing just this. I haven’t seen the frames in question but I know it can be done although, by the time you’ve finished, you would have wished you hadn’t. It’s a very hard problem in computer vision and theres not magic computer program you can just stuff it in for it to be done in.

7
      • Find a local art-supply store, and tell them you want to sculpt a Maltese Facon out of clay. Ask for something that stays workable until it’s oven-cured, and don’t forget to ask about what kinds if finishes you can use on it to get the metallic look. Unless you really don’t want to do it yourself. In that case…
  • Forget all that scanning of movie frames/3-D computer modeling. In particular: molding anything wouldn’t be worth the cost and bother unless you could get a cast off the original, and that isn’t possible. Find a college art instructor or a local artist who works in clay to make you one. Sometimes big art stores run classes, and that instructor is an option as well–ask at the art store if they know anyone who could do it. …If someone is well-practiced at sculpting, it will take them maybe all of a day or two of work to do it darn near perfect, from just the photos you have now.
    ~
8

Remember to make it heavy. It is remarkable how much heft adds to the realism of something like this.

9

As the one in the movie turned out to be “lead, lead, its made of lead,” you’ll need to get a lead ingot and a REALLY hot fire…and Sydney Greenstreet to have a fit when you’re finished with it.

10

So you guys who are suggesting the old-fashioned way, is that because you are in a position to say, like Shalmanese, that it would be hard/impossible to do with computers? Or are you just WAGing?

Shalmanese: can you point to any sites or published material that deals with this problem and its complexities? I was hoping that since the technology to make 3D objects from computer models (the second half of this problem) is getting more advanced all the time, that similar advances were being made on the first half. I can see that scanning from a series of 2D images is completely different from, say, laser scanning a real 3D object, but I was hoping some progress was being made.

Ultimately, of course, this is really a hypothetical exercise, since unless someone had said, “Sure, buy DingusScan 4.0 and an Epson 3D printer at Best Buy for $200, and you’re all set,” I probably wasn’t going to follow through. Just curious as to how difficult and expensive it really would be.

11

Sure, the general problem is called 3D reconstruction and you can google for that and get lots of papers.

Hartley’s Multiple View Geometry is the standard reference in the field and the math is not unbearably bad in that one.

Nister2004 contains a compact overview of the field although I think that he has a few glaring omissions and is rather selective in the techniques he chooses to cover.

Andrew Davison has lots of pretty videos on his site but he doesn’t focus much on the reconstruction side.

Be warned though, the results you see in academic papers typically are not representative. Theres a tendancy to cherry pick your most impressive test runs to show the world and ignore the rather glaring errors that might happen when something goes wrong. Nister is especially bad in this respect IMHO.

12

BTW: If you manage to find somewhere to download the digital version of the frames, then I could give you a much more certain answer as to how possible it is. I can tell you now that if anybody is moving in the background, it’s almost certainly a no as no algorithm I’ve seen even attempts to deal with that case.

13

Thanks! I’ll look them over.

I forgot to mention that this link shows a series of the frames in question.

14
      • The art-sculptor way would be lots easier and cost far less, and would end up doing about as well as the computerized route. “Artists” tend to charge a lot more for their time than “art teachers”, however. And just where do you plan to go to get molds made from 3-D computer files? I’ve no doubt that there is somewhere that could do it, but it’s pretty likely to be extremely expensive.
  • As for the movie one being made of lead, also I’d say forget that right now. Assuming you are willing to pay the several thousand dollars to have one cast from any kind of metal, they’re going to use bronze, and if you ask about lead they’ll say they can’t for health reasons and that their experience is in bronze and you need to go somewhere else if you want lead. The only large-scale casting done is in bronze, because it works best. And if you say that you have some 3-D computer files, they’re going to chuckle and ask if you have any pictures of it, and they will just work off them anyway.
    ~
15

There is software out there to build a 3D model from pictures by taking points you pair up and doing the trigonometry to relate their 3D positions. (I saw a working demo of one such program years ago. Cool!) This one costs $895; I suppose others have similar prices. This site lists a bunch, many in the ‘Close Range Photogrammetry’ section. Maybe you can find a demo version of something.

Making it physical is left as an exercise for the reader.

16

You see, this is exactly what I wanted, and after someone else said authoritatively that it wasn’t possible/practical.

And not two minutes after I bought one of the cheap knockoffs! Literally.

Well, I wasn’t going to spend $900 for just the first step, anyway. Of course, if I could find someone who already had the software, or who provided this as a service…

Thanks rjk.

17

If you take a look at those links, they promise to do different things and neither is what you want. The first link, you have to manually specify the correlations between the different pictures and then it will make lines joining those points together to form planes. Thats fine if your modelling something boxy like a car with relatively few, flat surfaces. But if you want to do something like that for a maltese falcon, every corner on the falcon has to be matched. And curved surfaces are impossible. This would not only be a) mind numbingly tedious but also b) likely to be reasonably inaccurate since a slip of just a few pixels would give you bad matchings. The holy grail of 3D reconstruction is to have all this happen as close to automatically as possible.

The second link applies only to specialised non-visual hardware such as laser scanners. This is fine if you already HAVE a maltese falcon that you can measure, but if all you have is video frames, then none of those resources is going to help you a bit.

18

Ah, I see. I thought that if you joined a few key points, the program would figure out the rest. The site’s FAQ mentions that “organic” forms are harder to do, but stops short of saying they’re impossible.

Thanks for clearing that up Shalmanese.

19

After looking at the frames in question more closely, I would say the shininess of it is going to be a major headache for any automated system. AFAIK, nobodys even started on reconstruction for anything but lambertian surfaces (ie: surfaces that look the exact same colour no matter what angle you look at it from). It would be interesting to figure out if you could do it or not given a known light source but your looking at a 3 year PhD project, not off the shelf software.

20

There is no automatic way to achieve what you want.

Get an artist to either make one out of clay that you can mould from, or if you really want a 3D one first (for whatever reason) then find a 3D artist who’ll help.

Really, though, you’re asking the impossible.