Project Overview
The Company:
Swiss analytics giant that provides state-of-the-art waste assessment solutions
Industry:
IT & Analytics
Services:
Data Annotation
Business Needs
Accurately label all interest areas in kitchen waste images and food images for training machine learning models of the client so that the models can interpret visual information like humans.
The Challenges
- Hiring and training data annotation specialists with experience in food and beverages space
- Accurately recognizing food images with multiple food items
- Understanding, working and labeling a wide range of diverse European food products
Solutions and Results
Want quality image annotated datasets for your ML models?
Speak to us today- We followed a multi-step annotation process comprising segmentation, labelling, audit, and review. A web-based tool enabled secure login to client’s portal.
- Garbage throwing process’s visual data was iteratively analyzed. Interest items in the images were reviewed from various angles.
- Images were segmented and tagged product list-wise and as per standards followed by different food outlets.
- An audit mechanism was implemented, involving the client conducting a parallel image labeling activity in-house.
- Fuzzy or low confidence images were routed for separate validation. While re-labelling was initiated for erroneously labeled or missed out images.
- Training data was automatically fed to machine learning models that rapidly matured them.
- Real-time intelligence on important metrics like food wastage type, wastage ratio etc. significantly decreased the food wastage.
Business Impact
100% accurate dataset of annotated images
Accelerated food waste analytics
Multi-metric guided actionable insights
