Not reinventing the (kart)wheel

Time is of the essence when there is undetected bananas about. We will probably spend some time training our algorithm to detect all sorts of potential road impediments, but in order to rush to market and deliver value as early as possible, we instead grab a mushroom and cut some corners. Much like Boo in Mario Party, we can steal other people’s hard work!

We started by forking the Tensorflow repo to make the basis of our app. Tensorflow is an open-source end-to-end platform for machine learning, and a great starting point for our app. We can use it to create models for machine learning. And while speaking of thieving, Tensorflow is also built on other public packages – also a form of stealing.

Speaking of models, why make your own when there are several detection models we can “borrow”

Mwa ha ha! Surprise, Luigi! It’s-a me, King Boo!

King boo – Luigi’s mansion 3

Others have also created several models for mobile use that we can reuse while we try to train our own. MobileNet model is one of these and can be used straight out of the box. These models are in turn also trained on public data that we can “steal” and modify as we please.

We also “borrow” more tangible things. Like non-licensed photos of bananas for training our detection model and map and geo-services for location data.

TL;DR Thieving is good – at least in programming

2 thoughts on “Not reinventing the (kart)wheel

  1. Interesting! What if it’s a straight banana in the road? Is there a certain degree of bending that must be met for it to be classified/recognized as a banana?

  2. Excellent question! We are currently creating a dataset to train our own model, and we’ll be sure to include bananas of all shapes and forms. Expect to see us move around outside tomorrow during daylight with bananas and banana-like objects!

Comments are closed.