The idea

Our teammate Ferja loves coins so when we arrived in the UK he exchanged some euros to pounds and after buying a snack it was hard to tell how much he had left. So we thought about a web app, compatible with any device, that tells you how many coins of a foreign currency you have and their value in your currency, by simply taking a picture.

The development process

We wanted to do it with some Computer Vision API and chose one called Clarifai from several ones like Google Cloud and Microsoft Azure. We made a lot of pictures of coins on a hand and hands without coins. Then we trained the model in the Explorer UI that Clarifai offers and we accomplished to distinguish between 1 euro and 50 cents coins, but not in all cases.

The next step was sending or images and telling how many coins of each type it had, but we failed. We realized that the API only tells us whether there are coins or no, not how many of them, and we couldn't find another easy to use API that did that. In addition, we needed to set up a server with Node.JS and send the images via HTTP forms, but only one of the team knew some HTTPS and JS. By 10 hours left, we realized that what we wanted is more complex than what we thought we could do in 24h.

The plan B

For all of that, we change our minds completely and we focus on the none Computer Vision aspect of the web app and preparing it for an easy upgrade to what we wanted at the beginning. We implemented a manual input instead of an automatic one: you drag the coins (euros) you have to the bottom part of the app (representing your wallet) and also see how much do their worth in pounds. The drag and drop feature took us a lot of time because it uses jQuery and we don't know how it works and also we wanted to make it usable from a phone including an extra library.

What we ended up with

Finally, we have done the frame for the final app and half the core of the computer vision part. But for time restriction we implemented a manual selection of the coins instead of an automatic one.

Built With

Share this project: