+
Update: We built the Inside/Outside Project for the event. More Information.
Description
Use TensorFlow and the wq framework to create custom image recognition applications for citizen science! A project idea for Create Together Day at the 2017 Citizen Science Conference.
Create and use a wq-powered mobile app to take photographs, or search for images online.
Use our pre-installed TensorFlow image, or experiment with your own installation.
Deploy the neural network in your own app, or just test it with this one.
Project Idea by:
- S. Andrew Sheppard (Houston Engineering, Inc.; Computer Science at UMN)
- Marco Willi (School of Physics & Astronomy at UMN)
Event Logistics:
Create Together Day on May 17, 2017 from 8:45am - 4:30pm at Ralph Rapson Hall, UMN, 89 Church St. SE, Minneapolis.
What's Provided
We will be using and demonstrating the following open source libraries:
We will provide cloud servers with all of the necessary software preinstalled for use at the event. No experience with any of these technologies is required to participate, but you are welcome to experiment with your own copy.
What to Bring
Bring your project ideas and your smartphone/tablet (Android, iOS) or laptop. You may want to install the heigeo App Showcase native app.
Detailed Plan
Before lunch, we will form 1-3 teams and each define an image classification project. The ideal project is something that can be photographed during the event. If you are more interested in something that isn't locally available, you can certainly use existing images from the web or other sources.
Some example project ideas include:
- Categorize the tree species at the UMN campus
- Document the utilization of bike racks.
- Train a general-purpose classifier for items likely to be found at a University.
With the projects defined, we will create 1-3 simple apps using the wq start tool. When we are done, each app should have (at a minimum):
- a simple form for defining the classes
- a form for capturing, geotagging, and manually classifying training photos
- a map to view collected photos (assuming the photos are geotagged)
After lunch, we will work on gathering photos to train and improve the classifiers. This will be the most interactive part. Depending on the project ideas (and the weather), we will try out go out into the University campus to take a few dozen photos. We will then come back and see how much the classifications improve with the new data. If there is still time at the end, we will install the TensorFlow plugin into each of the apps we built in the morning. The goal is to have apps that automatically improve their ability to classify images as more photos are submitted.
Back to Home