Bringing real world objects in TDM

[Outlooker]

I read here somewhere that people are photopgraphing real world objects to produce textures from them.

 

Now there is a game in development that makes much use of photopgraphing real world objects and with some automation bringing a whole model (geometry+textures) at once in the game.

 

I wondered if that kind of thing could also be done for TDM?

 

Here they give an example:

 

http://www.theastron...g-ethan-carter/

 

 

 

It uses some hand work on the model, but in the attached pics is the start and end product of the procedure.

 

They use it for small things like stones up to things like churches.

 

 

Might some TDMers start not just making pictures to produce textures, but import real world objects - full of detail - at once, and releatively quick, at least much faster than building complex things manually?

[Springheel]

If you read the article, you can see that it's full of descriptions about how difficult this technique is to use:

 

 

I could go on for hours about the challenges of acquiring imagery for photoscanning, but don’t be fooled into thinking that it ends there, with “simply” getting perfect photos.

 

You often need to mask out and process photos in a number of ways. Sometimes you need to manually help align photos when the software gets confused, and then, fingers crossed – software will finally spit out an object. As you have seen in our raw samples above, there might still be errors and gaps that you need to fix manually. Then you either accept okay texture quality or begin long, tricky and tedious work of improving the texture projection and removing the excessive lighting and shadowing information to achieve perfect looking source object.

 

Job done? Still far from it.

 

Scanned object will usually weigh between 2 and 20 million triangles. That is the entire game’s polygon budget, in a single asset. You need to be really skilled at geometry optimization and retopology to create relatively low poly mesh that will carry over most of original scan geometry fidelity. Same thing with the texture – you need amazingly tight texture coordinates (UVs) to maximize the percentage of used space within the texture.

[demagogue]

I think if people invested the time to learn how to model, they could become faster at it, and after a while it'd be second nature to quickly put together some new objects for their FM. That'd be the ideal IMO. But the learning curve is still tough. I used to put a lot of effort into learning how to model and I still feel very slow & awkward at it.

[rich_is_bored]

I don't believe it's possible to automate optimization. Somewhere in the pipeline you need a knowledgeable person to clean the object up. Does the finished object have good topology? Is it's polygon count reasonable? Would it benefit from a shadow mesh? How are it's UVs packed? Is the texture resolution reasonable? And if the object is especially large there are cases where splitting the object into pieces can be beneficial.

 

This could be useful in some rare cases but only for those with modeling experience. Without it you have beautiful models and horrible performance.

[Arcturus]

I did some tests with open source PPT GUI. This is my head scanned:

 

http://youtu.be/CxrP1TcdqDc

 

What others said: it requires cleaning up and retopologizing (there are some automatic and semi automatic tools that can help). Also these guys are using commercial software which is probably a lot better than open source.

[SteveL]

Fascinating. I'd idly wondered if it was possible to generate a 3d model out of 2d images, but didn't even know what to google for to ask the question!

[jaxa]

Here's what you need:

 

http://www.theregister.co.uk/2014/07/22/boffins_fill_a_dome_with_480_cameras_for_3d_motion_capture/

[SteveL]

Intriguing to see how that works but setting up a 50-foot geodesic panoptic dome with 500 cameras is a bit beyond me. Is there a budget way to play with the technology, albeit with less impressive results?

[jaxa]

2 kinects, a raspberry pi, and a pack of chewing gum?

[STiFU]

It's basically "structure from motion". Google that and you'll find loads of information. If you have a sequence of images, at first the camera trajectory is estimated. With the relative camera positions now known, you can analyse the parallax of corresponding points in the images and retrieve their depth that way. The link jaxa posted is the easy way of doing structure from motion, because the cameras are completely calibrated, meaning that their positions and intrinsic parameters are already known and don't need to be estimated.

 

Oblivion used the same technique for capturing faces by the way.

[SteveL]

Excellent, thanks. Only had a quick scout so far but I see it *can* be played with for free