Virtual photogrammetry test

[bigben]

Another academic exercise in photogrammetry. I'm sure someone will recognise the source, if not the actual building. (No I can't share the object) The trees are XFrog.

[Dune]

Cool exercise. Is it a real house that's been 'photogrammed' or a small model? It works quite well, but lacks small detail, unfortunately. If you add bump, it would enhance it a lot, but I guess it's hard to make detailed bumpmaps from the photogram (or whatever you call the base).

[bigben]

In game screengrabs.
http://elderscrolls.wikia.com/wiki/Honningbrew_Meadery

The UVs aren't split by material as the originals would be so it would be difficult to do more than hack a few bumps maps based on the colour of the texture map.

[TheBadger]

In game screengrabs.

HA HAHHA HAHHA HAHHA HA HA HAHHHAH AHHA! 

F-ing brilliant! That is just too good. Yes I like this. Very funny and smart.

[bigben]

Bigger, and yet smaller.

[Kadri]

Nice tests!
I immediately thought about Skyrim in the first image but the use of "actual building" threw me off
Can you show wireframe images of the objects and how big are the file sizes Ben ?

[bigben]

Waiting for a train at the moment...  First one was about 300k faces, second one 900k faces but could really use more. Point cloud was 5million points for the second.

[Kadri]

Thanks Ben.

[inkydigit]

Excellent stuff Ben!

[bigben]

The mesh of the first one

[Kadri]

Thanks Ben. Looks quite dense.

[TheBadger]

Thanks Ben. Looks quite dense.

Yeah, for that amount of polygons, I would expect to see much better detail... Even down to a rain drop on the side of the roof. The amount of geometry vs the resulting quality remains an issue with this method of creation.

I guess the question for me now is, if you take a person and make him/her an expert modeler, and also an expert at the photogrammetry stuff, would he/she be faster with the one or the other method? I guess, my interests in this remain primarily in the practical. But I have to admit the tech is pretty fascinating over all.

What are your thoughts Ben? Im guessing that regardless of what the reality is now, that this tech will advance very fast. I do feel that it is important to understand it now while it is still pretty young. In another year or three, playing catch up with understanding everything will likely be a large amount of work in it own, not to mention learning how to apply the tech.

Of course I have seen rather impressive applications of this stuff already. But it dose feel not quite ready for the average desktop user?

[mhaze]

I'm fascinated by this, clever work, the second castle is very convincing.

[bigben]

It's quite dense because that's the way photogrammetry works.  I simplified the first model a little in Meshlab before texturing in Photoscan, and could have done more simplification, but the need/incentive to do so is lacking.  You have to remember that these are derived from gaming models so the original models aren't going to have a complex structure to start with. The game art relies on good texturing to compensate for the lower poly count..  But photogrammetry is going to produce a relatively even mesh over the entire surface area, even if it's completely flat.

I did a quick test to see if I could recreate my character but the lighting was a bit harsh. Had a bit of a think about the problem before realising the bleeding obvious... my character can cast a "fill light" spell before I freeze the game for the screen grabs    Just find a room of the appropriate size and cast mage light around the walls to provide studio lighting.

Veering slightly off topic... it is proving to be a valuable testing ground for shooting scenarios and their effectiveness for photogrammetry.  Don't have to worry about lens distortion, motion blur etc... and there's plenty of test subjects in various states of ruin. With improvements in digital cameras and computer processing power making this technique more practical, there is a growing need for visualising the resulting data and TG is possibly one of the simpler options for this.

[bobbystahr]

It's quite dense because that's the way photogrammetry works.  I simplified the first model a little in Meshlab before texturing in Photoscan, and could have done more simplification, but the need/incentive to do so is lacking.  You have to remember that these are derived from gaming models so the original models aren't going to have a complex structure to start with. The game art relies on good texturing to compensate for the lower poly count..  But photogrammetry is going to produce a relatively even mesh over the entire surface area, even if it's completely flat.

I did a quick test to see if I could recreate my character but the lighting was a bit harsh. Had a bit of a think about the problem before realising the bleeding obvious... my character can cast a "fill light" spell before I freeze the game for the screen grabs    Just find a room of the appropriate size and cast mage light around the walls to provide studio lighting.

Veering slightly off topic... it is proving to be a valuable testing ground for shooting scenarios and their effectiveness for photogrammetry.  Don't have to worry about lens distortion, motion blur etc... and there's plenty of test subjects in various states of ruin. With improvements in digital cameras and computer processing power making this technique more practical, there is a growing need for visualising the resulting data and TG is possibly one of the simpler options for this.

brilliant ground breaking, in a way, stuff Ben...totally beyond my ability to reproduce but a treat to follow here.