Image Annotation for Swiss Food Waste Assessment Solution Provider

Project Evaluation - The Scope of work included mapping images to the client's repository, tagging and segmenting based on standards, and implementing a validation process for uncertain annotations.

Preparation - Before starting the annotation, we arranged for annotator training, prepared annotation guidelines, developed consensus processes, and aligned workflows with the clients software and tools.

Input Data - The client provided food product images via their software. Our team of annotators assessed these images and meticulously planned a workflow for efficient image labeling.

The Process - The images were annotated through a three-step process.

1. Labeling against existing repository

• Labeled food garbage images using a drop-down menu within the client's portal.
• Cross-referenced discarded food with a predefined list of food items.
• Reviewed images from various angles for accurate identification.
• Ensured precise matching with the existing database for consistency.

2. Tagging and segmentation

• Tagged and segmented images according to the client's product list and quality standards.
• Categorized and outlined food items to conform to defined standards.
• Applied annotation types (bounding boxes, polygons) to highlight areas of interest.
• Assigned appropriate labels or class names for accurate categorization.

3. Handling low confidence images

• Flagged images with uncertain or fuzzy identification.
• Handled flagged images separately for a thorough review.
• Sent uncertain images to the client for additional validation.
• Ensured the highest accuracy and reliability through meticulous validation.

Quality Control and Validation

• Implemented a multifaceted quality control process for high data accuracy.
• Parallel labeling was conducted by the client's in-house team for cross-verification.
• Re-labeled data based on client feedback to ensure accuracy.
• Adhered to rigorous standards and guidelines throughout the annotation process.

Client Review, Feedback, and Monitoring - We maintained ongoing collaboration with regular updates and clarifications to keep the project on track. Working with the client’s in-house team for continuous validation, we received feedback during governance calls to ensure alignment with expectations.

Final Deliverables - Labeled food images were automatically uploaded to the client's portal for machine training, with reports and dashboards providing insights into the quantity and categories of annotated images.

Technology Utilized - Secure login to the client’s portal through a web browser, ensuring data security and integrity throughout the process.