http://mi.eng.cam.ac.uk/~qp202/my_papers/BMVC09/
-- 
Darren New, San Diego CA, USA (PST)
   You know the kamikaze monsters in Serious Sam
     with the bombs for hands, that go AAAAAHHHHHHHH!
   I want that for a ring tone.

Post a reply to this message

Doesn't appear to be very accurate though... (And that's only for 
objects with trivial geometry.)

Post a reply to this message

Invisible wrote:
> Doesn't appear to be very accurate though... (And that's only for 
> objects with trivial geometry.)

Well, it *is* a doctoral thesis. Wait till it gets productized. You're 
seeing a proof of concept here.

-- 
Darren New, San Diego CA, USA (PST)
   Human nature dictates that toothpaste tubes spend
   much longer being almost empty than almost full.

Post a reply to this message

>> Doesn't appear to be very accurate though... (And that's only for 
>> objects with trivial geometry.)
> 
> Well, it *is* a doctoral thesis. Wait till it gets productized. You're 
> seeing a proof of concept here.

LOL. Is "productized" even a word?

Anyway, you're acting like once you can turn a pair of images into a 
grainy, vaguely-correct 3D model, it will then be a "trivial" task to 
turn this into something that actually works properly. I'm not sure it is.

-- 
http://blog.orphi.me.uk/
http://www.zazzle.com/MathematicalOrchid*

Post a reply to this message

Orchid XP v8 wrote:
>>> Doesn't appear to be very accurate though... (And that's only for 
>>> objects with trivial geometry.)
>>
>> Well, it *is* a doctoral thesis. Wait till it gets productized. You're 
>> seeing a proof of concept here.
> 
> LOL. Is "productized" even a word?

It is in my country, yes.

> Anyway, you're acting like once you can turn a pair of images into a 
> grainy, vaguely-correct 3D model, it will then be a "trivial" task to 
> turn this into something that actually works properly. I'm not sure it is.

I imagine one could not too difficultly add code that would help the 
software figure out what it needs to see to be correct. It'll certainly be 
easier to start with the almost-right model than to start from scratch.

-- 
Darren New, San Diego CA, USA (PST)
   Human nature dictates that toothpaste tubes spend
   much longer being almost empty than almost full.

Post a reply to this message

> I imagine one could not too difficultly add code that would help the 
> software figure out what it needs to see to be correct. It'll certainly be 
> easier to start with the almost-right model than to start from scratch.

Even if it can't get it 100% correct, it would be great for very quickly 
getting a 95% correct mesh done (with textures!), that only requires a bit 
of tidying up in a modeller knowing that the overall dimensions and 
proportions are correct.

Post a reply to this message

>> I imagine one could not too difficultly add code that would help the 
>> software figure out what it needs to see to be correct. It'll 
>> certainly be easier to start with the almost-right model than to start 
>> from scratch.
> 
> Even if it can't get it 100% correct, it would be great for very quickly 
> getting a 95% correct mesh done (with textures!), that only requires a 
> bit of tidying up in a modeller knowing that the overall dimensions and 
> proportions are correct.

I guess I was hoping for a system that would take a pair of photos and 
reproduce the blades of the grass, or the bumpmap of the grain in the 
wood, or the individual sugar grains in a sugar jar, or similar. You 
know, like the human eye can. The demo here seems very far from this level.

Post a reply to this message

> I guess I was hoping for a system that would take a pair of photos and 
> reproduce the blades of the grass, or the bumpmap of the grain in the 
> wood, or the individual sugar grains in a sugar jar, or similar. You know, 
> like the human eye can.

That's impossible from two photos, you may *think* your eyes can do this 
perfectly, but a lot of it is just your brain making stuff up based on what 
it is expecting to be there.

Post a reply to this message

>> I guess I was hoping for a system that would take a pair of photos and 
>> reproduce the blades of the grass, or the bumpmap of the grain in the 
>> wood, or the individual sugar grains in a sugar jar, or similar. You 
>> know, like the human eye can.
> 
> That's impossible from two photos, you may *think* your eyes can do this 
> perfectly, but a lot of it is just your brain making stuff up based on 
> what it is expecting to be there.

Wouldn't it depend on the resolution of the images and the size of the 
objects in question?

You can't make out sand grains on the brach from 4 miles out to sea, but 
I'm damned well sure you can when you're laying in the sand reading a 
book. :-P

Post a reply to this message

> Wouldn't it depend on the resolution of the images and the size of the 
> objects in question?

No, it's a simple fact that with only two photos there is never enough 
information to make a 3D model and be sure it is correct.  Your brain always 
has to add in unseen or assumed details, which most of the time are 
believably correct, but I guess this is extremely hard for a computer to do 
(as it's based on your experiences etc).

Using your grass example, I'm sure lots of blades will be hidden by others 
in the view from one eye, but plainly visible in the view from the other 
eye - how can a computer know what to do with that?  Even if the computer 
could somehow match up individual blades of grass between the two images, 
the partially hidden ones it would just have to guess where they were within 
a certain range of positions.

Post a reply to this message