Inspiration
There must be sometime when you see some nice pictures that evoke your desire to find out the photograph location, so you can visit it in person sometime and get your own shot. However, a lot of pictures online may lost their GPS infomation, and you don't even know the place name to google it. This inspires us to develop an app that can geo-locate any photo, bassed on picture content, using deep learning.
What it does
This is an web app, where you can upload any picture you want to know the location where it was taken. And then geolocate it on map. But, currently attribute to time and data limitation, we just focus on street art pictures in Los Angeles.
How we built it
We train a Convolutional Neural Network model with LA street art pictures. Then if we feed any other LA street art picture you find to the model, it will recongnize the art from picture, and return infomation help geotag it. This is our main engine. Then we build the web app implementing ArcGIS Javascript API, to geolocate the photo.
Challenges we ran into
- Find appropriate data to train
- Build and train the model
Accomplishments that we're proud of
We're able to return expected result using our model.
What we learned
Learned the architecture of CNN model
What's next for Street Art Hunter
Cover more wide area for outdoor pictures
Built With
- arcgis-javascript-api
- deep-learning
- javascript
- python
- tensorflow
Log in or sign up for Devpost to join the conversation.