- Vision is a software, not a hardware problem. Every vision effort I see has more hardware than software people. The first thing one of the Darpa Grand Challenge teams did was to buy a turbo Hummer H2, beefed up the suspension, installed every sensor they could imagine, and then spent the next year hooking them all up. Never did they really get to the vision aspect.
- The codebase is a few hundred thousand lines of code, excluding OCR
- The first free codebase that gains the critical mass will win.
- There is no free vision codebase with critical mass.
- There is no free driving engine with critical mass.
- There are plenty of free codebases to use.
- The existing free (but non-critical mass) codebases is why on could really put 5 developers to work today, and 20 developers in 3 months.
- The biggest competitor today is someone with code but who hasn't made it free yet. Note, their codebase is probably hacked together because they were in a rush. A company can work a little more diligently and produce code that is easily usable by millions of people. Netscape's code had issues after it was released and it took years for a community to rally around it because no one understood it and it was a mess.
- There are only a few people who have something even close to usable to drive cars.
- Most people even at Microsoft and Intel think this idea is too far out.
- We know lots of smart people.
- A goldrush is about to start. One reason not to give you my book is that it would just distract you. I've given all you need to know.
- Neural networks and computer vision are not magic. We will be able to take the workings of your eye and convert it to software one day. It will be small enough to run very well on today's PC, even without leveraging the parallel processing of the GPU.
- We only have to solve the software problem. If you have a car driving down the highway and get broadcast on the Channel 7 news, the rest of the problems will solve themselves.