Inspired by zooniverse as a way for community to participate in building the best datasets for the future needs of machine learning.
What it does
It does what zooniverse does not! It allows multiple users to simultaneously add localized (positioned) bounding boxes around objects (or people in this case). It allows for configurable labels (just place in your set of labels with your custom colors and you can start labeling photos).
The backend allows users to upload images, retrieve images, attach classification/object localizations to images, and retrieve those classifications.
How I built it
backend: node.js frontend: React because it's awesome, React-Konva for canvas and modified drawing of rectangles, Redux for state DBs: Mongo Atlas on AWS for photo storage, PostgreSQL on Amazon RDS for user content
Challenges I ran into
I'm just one person and I'm very slow at designing sites.
Accomplishments that I'm proud of
I think this will be a useful tool that organizations could rely on once it is more resilient. Everyone will need better datasets with special labels. Supervised learning is still an extremely valuable tool and the data is paramount.
What I learned
Canvas is tricky, but Konva is a nice library. Mongo Atlas is a very nice free service for development since it is available in many of AWS' regions.
What's next for HackernotHacker
- Fix node serving of static site
- Allow for uploadable labels
- Set up point system
- Allow consensus mechanisms to vote on best set of object localizations
- Remove hackernothacker labeling and make a