Inspiration
Walking through an unknown city, have you ever wondered which building you see in front of you? Searching on Google Maps just provides you a name and reviews about places of interest, and then you have to google it. But what if you could just take a photo and get all the necessary information and links just in your smartphone?
What it does
Using ML and computer vision, Find Your Place to Visit recognizes an uploaded photo and matches it to a known database of places of interest in a particular city.
How we built it
We crawled a custom dataset of Barcelona's 29 most popular tourists' places from Google Images. We trained a machine learning model to recognize these sights and, with the help of serverless computing by AWS API Gateway and Lambda, connected it to a mobile-targeted web page which allows to upload an image and see some information about it.
Challenges we ran into
Crawling pictures from Google Images was a bit challenging because that it was necessary to build a precise and diverse dataset and without any duplicates. Also we ran into some troubles with AWS Serverless Computing, dependencies versioning and memory usage limits.
Accomplishments that we're proud of
We managed to collect a relatively big dataset on very noisy data such as output of Google Images search. We trained a model with acceptable accuracy for not a very common task. And finally we connected everything in a user-friendly mobile web application.
What we learned
We learned useful things about new (for us) technologies and APIs, we learned how to work with such innovative thing as serverless computing.
It was our first time
...to use serverless computing, to collect an image dataset, to collect good data corpus.
What's next for Find Your Place to Visit
We are planning to extend our list of cities, to improve the quality of prediction and to build a desktop-compatible version of the application.
Built With
- amazon-web-services
- google-maps
- imgaug
- keras
- python
- scrapy
- tensorflow
- vue

Log in or sign up for Devpost to join the conversation.