I made a website where you put sunglasses on a photo - seemewith.com. After face.com shut down its facial landmarking service I deleted it, but it was a failure anyway. I spent a ridiculous amount of time implementing various strategies for turning photos of sunglasses into 3d models (using bundle adjustment, advanced appearance models, and some symmetry specific photogrammetry) - in the end I did the simplest thing I could think of and it worked better than any complex option. The results were still a bit poor though.
Another side project was a flash based plugin to do teeth whitening on photos. End up implementing a decision tree approach that worked ok, but it didn't seem to really have legs as an idea.
What else... I spent an inordinate amount of time on the Netflix prize, that really amounted to nothing - although it was fun to try a lot of different approaches at a tricky scale.
I can understand that that app does not fly; recognizing faces and showing matching animals outsode the original photo just does not look too enticing.
I can think of various ways to, IMO, make this an interesting app:
- put the heads in the photo, and tween a movie between the two (probably fairly hard because you will have to reliably cut out the background)
- use the camera to record video, animate the animal head in sync with what the head does (ideally, adjust the voice sound; more bass for a bear, twitter for a duck, etc) (probably not too hard to do)
- make it a game: using the camera, "let's see who can make his face best like this elephant", with solo games for practice and for fun ("can you beat your record of ten faces in 30.34 seconds?) (technically simple, but it probably will be a lot of work to get the game 'right', collect funny animal photos that humans can learn to match, etc)
Such a small sampling of my many failures. I do try to celebrate them though.
Lessons learned? Well I did a full run down of all the projects I could think of (that got somewhere at least) a little while ago and I did have some interesting insights. Mostly I try to invent new categories of things and fail (both the idea and the newness thereof tend to fail in case you're wondering).
The promising direction that I hadn't followed was my first ever paper, and I think thats because it was a Human Computer Interaction project. At least for this project the focus was on new ways of doing tasks that people already do with computers - so it had some built in relevance. So now I'm trying work with my own psychology and find innovation in interaction - not in the entire category of product.
Another side project was a flash based plugin to do teeth whitening on photos. End up implementing a decision tree approach that worked ok, but it didn't seem to really have legs as an idea.
What else... I spent an inordinate amount of time on the Netflix prize, that really amounted to nothing - although it was fun to try a lot of different approaches at a tricky scale.
Just released 'What animal?' on the App Store, that's pretty much doing nothing download wise - http://scottvallanceapps.com/app/what-animal-are-you/