LOL Vision
Build detalied Image datasets for multiple Machine Vision applications.
Land O' Lakes Inc.
Arden Hills, Minnesota (US)
Food & AgricultureLinks
The mobile apps are distributed internally
Problem
- No public dataset available for custom Computer Vision applications
- Even if open-source images available, difficult to mimic actual use case conditions
- Difficult to find free use images for commercial applications
- Difficult to control quality of open-source images thus limiting the training data
- Client had Computer Vision ideas for wide variety of animal species. Again, difficult to find images for uncommon species.
- Client had some images with them but without annotations
- Client wanted to collect some metadata along with images as well for various ideas
- Clicking fresh images possible. But cumbersome for users to transfer to Cloud along with metadata
- Internet connectivity an issue in remote farms where most images would be clicked
Solution
- Build a mobile app with offline image capturing
- Sync images when device online
- Add metadata while browsing images one by one
- Implemented Machine Learning based Image Acceptance Algorithm to control quality of input images
- Web app to allow users to upload old images
- Web and mobile apps allowed easy collaboration leading to large image datasets very quickly
- Use same apps for multiple species. Switch seamlessly between species
