Microsoft Photosynth – A super mockup?
Wednesday 20 June 2007 - Filed under News
Photosynth form Microsoft’s lab looks really amazing. The program analyzes a group of photos and joins them together forming a 3D model view. But it’s too good to be true.
Based on their “How do you do it” page, it says that the program searches through images for distinctive features, corners, doors etc. If the same feature is found in multiple images, the program will figure out the images’ 3D positions. From the demo we can see that, images are rotated/skewed/scaled in order to fit into the 3D model. So, the question is how the program knows how to rotate/skew/scale images so that it produces the best matching result. It is quite an NP-Complete problem for even just 2 images, let along hundreds of images. Very likely, the program needs to adopt some artificial intelligence algorithm to guess out a good solution (may not be the best). It’s quite impossible for the program to come out a seamless 3D image viewer like the demos.
As conclusion, I suspect the preview demos have some or lots of human touch ups instead of purely computer generated and the final product will be quite far away from what we see from the demos.
2007-06-20 » 5566
20 June 2007 @ 7:12 pm
Fantastic “Microsoft Photosynth…
When was the last time you’ve got excited by a Microsoft application? Maybe you will say “today” after watching the video below…
14 August 2007 @ 2:25 am
Have you seen the programs that generate panoramas automatically?
I once got up in a tower and shot 103 photos – different angles, zoom levels and so on. Basically I was just playing. Then I imported all of them in a panorama software and hit Generate. It resulted in a beautiful 360 panorama.
Then, there are the motion tracking software. You give them a video and they figure out the 3D scene by themselves, generating 3D vertices and camera paths, so you can integrate 3D generated stuff in existing videos.
Have a look at http://www.realviz.com – they are not even the best in the field.