Photogrammetry

Photogrammetry is capturing real world objects – often with still images or video – and turning them into 3d objects. It’s amazing technology and has come really far in just a couple of years.

So far, in fact, that there’s free software available for anyone to use (though saying it’s finnicky to use is an understatement).

Here are a few tests I’ve been working on, to see what’s possible.

This was captured on a rainy day next to a field, using a total of 87 separate photos. It was run through Meshroom (many times) and finally imported into Blender. Total tris: ~1.6 million (~17 000 in remade low poly version). Render time ~10s per frame.

A big limitation of photogrammetry is that it captures the surfaces “as is”: they’re non-dynamic. There are no normal maps, height maps, specular, roughness and so on.

It took a while to recreate the slick wet look on the rocks (and excluding it properly on the moss). In the end, I decided to not create custom normal and specular maps, and instead found an effective way to use only the original texture.

Material nodes for the rocks and manhole cover

On the other hand, the scan of these mushrooms did not turn out very good.

Grass and detailed foliage will always come out bad in photogrammetry, so that’s no surprise. Sadly the mushrooms didn’t fare much better. But it will be an interesting challenge to try and clean them up, perhaps recreate that lovely subsurface scattering that mushrooms often have…