This is the third in a series of posts on bringing 3D scans to the browser. in this post, I'm getting very specific - looking at the difference between Autodesk's consumer scanning solution (123D Catch) and their new enterprise solution (Recap Photo).
To illustrate the differences, I'm using the same input images to both services. Then I'm running the resulting meshes through our Verold optimization tech for display in the browser. Going to keep this post pretty lean, and let the results do the talking.
The images used for the test came from a Scanathon at the Metropolitan Museum of Art. There are 27 images total, at 3648x2736. Here's a sample, so you can see what the sculpture looks like.
Running these through 123D Catch, you get a 3D model with 56K polygons. Here it is, unoptimized.
That loads pretty well, but to make it really snappy - useable in a web app or game - I turned on the optimization settings on upload. This brings the 123D Catch model down to 11K polys. This is in the realm of a character for a video game. We lose some quality, but relative to the original low-poly scan, it stays pretty close. Our optimization code did it's job well!
Now the fun part, running the exact same images through Recap Photo. With Recap Photo, we don't get the same cleanup tools provided in 123D Catch. Normally, I would have put the result through Blender, but for this test, since there's not that much extra environment data being dragged in, I left it as-is. The original Recap Photo model is much larger at 522K polys. This model loaded in Verold Studio, but it's definitely not suitable for realtime display.
So let's again use the Verold optimization tools to reduce the size of the mesh. For a first test, we'll decimate to 20%: this is still 100K polys (twice the size of the 123D Catch model).
And now 10% (~50K polys):
This was the best Apples to Apples comparison. We took the high res scan from Recap Photo, and decimated it down to the size of the unoptimized 123D model. You'll notice that despite our optimization, we still have the high fidelity detail that Recap Photo captured so beautifully. And now, to complete the test, we do one last run, this time optimizing to 3% of the original size, to 17K polys.
If you look closely, both 123D and Recap Photo left a big hole under the right side of the chair. As you might expect, there is a learning curve to taking photos that will result in quality scans. And it was lazy of me not to trim away the garbage polygons in the environment around the Recap Photo model. But for the purposes of this demo, I think the point is made. Very interested in feedback. What do you think? We're testing Recap Photo and 123D in ways they were never intended to be used - for real-time display on the web. And some of what you are seeing is the result of our engine and optimization tools. But I'm genuinely interested in feedback, what do you like and not like about what you've seen here.