Image Annotation for Swiss Food Waste Assessment Services

• Solution – We identified, categorized, and labeled thousands of food images from various hotels and restaurants, creating comprehensive training data for machine learning models.
• Input Data – The client provided food product images through their software. Our team of annotators evaluated these images and carefully designed a workflow for efficient image labeling.

Annotation Type Selection

To extract valuable insights from discarded food, we needed to identify and classify food items within cluttered scenes. Our data scientists assessed the food images and the required annotations to determine the best annotation types.

• Object Detection – Used bounding boxes to identify and localize individual food items within the waste.
• Image Segmentation – Created pixel-level masks to detail the shape and extent of food items.
• Classification – Labeled images to classify discarded food items by type.
• Key Point Annotation – Marked specific points or features on food items for analysis of dimensions and conditions.
• Sentiment Analysis – Annotated images with labels like spoiled or edible to assess the state of the food.
• Text Extraction – Extracted text information such as expiration dates or labels on packaged foods.
• Multi-Modal Annotation – Combined multiple annotation types to capture comprehensive details of food wastage.

Labeling Against Existing Repository

• Annotators utilized a comprehensive database of food item names to label the food images.
• This involved a meticulous process of cross-referencing the visual data with the predefined list to ensure accurate identification.
• Annotators reviewed each image from various angles and contexts to precisely match the food items with the corresponding entries in the database.

Handling Low Confidence Images

• Images that exhibited uncertainty were flagged for further validation to ensure they met the required standards of accuracy.
• Annotators consulted experts or used advanced tools to improve the confidence level of the annotations.

Quality Control

• A feedback-driven culture was fostered, and regular audits were conducted to ensure accuracy.
• Automated validation checks were used to maintain consistency, followed with expert oversight to oversee the process.
• The client conducted independent quality control by comparing our annotations with their own.

Deliverables

• Image Uploads – Food images were automatically uploaded to the client’s portal for machine training.
• Reports and Dashboards – Provided insights into the quantity and categories of annotated images.
• Advanced Analytics – Included metrics such as the day and time of food wastage, food types wasted, and the ratio of waste to prepared food.

Watch How We Revolutionized Image Annotation for a Swiss Food Waste Assessment Company