Berlin Computer Vision Group

Food Waste Estimation Hackathon - Computer Vision for Sustainability

Date: August 1st, 2025 | 10:00 AM - 8:00 PM
Location: Hasso Plattner Institute, Main building H, d-space
Prof.-Dr.-Helmert-Str. 2-3, D-14482 Potsdam

Event Overview

Join us for a hands-on computer vision hackathon focused on tackling food waste through image analysis. Participants will build systems that estimate food waste by analyzing post-meal images, contributing to sustainability efforts through machine learning.

Target Audience

• Data Scientists
• Machine Learning Engineers
• Computer vision programmers

Prerequisites: Participants must understand computer vision fundamentals. Familiarity with the open source dataset curation tool FiftyOne is highly beneficial but not required.

We recommend reviewing the documentation of FiftyOne and the following openHPI MOOC link for self-study preparation before the event:

* https://docs.voxel51.com/
* https://open.hpi.de/courses/computervision2025

The particularly relevant sections are neural network fundamentals, image-based regression, image augmentation, object detection, and image segmentation. Tutorial notebooks are available here:

https://github.com/andandandand/practical-computer-vision

Challenge: Food Waste Estimation from from Images

Build computer vision systems that analyze food images to estimate waste quantities.
Teams will work with post-meal photographs to develop accurate food waste measurement algorithms.

Technical Tasks
Object Detection and Segmentation

• Zero-shot food item segmentation with YOLO-E and YOLO-W and Grounding DINO + SAM

Dataset Creation and Management

• Image dataset exploration and quality assessment with FiftyOne
• Dataset curation using CVAT annotation tools for food images
• Custom segmentation dataset creation with food waste labels
• Validation set construction for before/after images
• Data augmentation techniques for diverse food scenarios

Food Waste Estimation Modeling

• Volume estimation algorithms (primary focus): Calculate remaining food quantities from segmented regions
• Waste percentage calculation: Compare before/after states to determine consumption ratios
• Multi-food item tracking: Handle plates with multiple food types
• Portion size standardization: Account for varying plate sizes

Advanced Features

• Synthetic food image generation for training enhancement
• Fine-tuning models for specific food categories

Rewards

1st Place: Amazon gift card with a value of 300 EUR
2nd Place: Amazon gift card with a value of 150 EUR
3rd Place: Amazon gift card with a value of 50 EUR

Scoring Rubric

Participants earn points based on implementation quality and food waste estimation accuracy:
Technical Implementation (50 points)

• Food Detection/Segmentation (25 points): Accurate identification and boundary detection of food items and use of model ensembles and manual annotation
• Waste Estimation Accuracy and Experiment Tracking (25 points): Precision in calculating food waste percentages and volumes, and use of appropriate metrics.

Best Practices Adherence (30 points)

• Dataset curation (30 points): Enrichment of the dataset through labeling, identification and resolution of leaky duplicates across train and validation splits, identification of uniqueness and representativeness of samples, richness of embeddings visualizations, data augmentation strategies, sharing of curated version of the dataset on HuggingFace
• Documentation and reproducibility (5 points): quality of public GitHub repository, inclusion of requirements.txt file and / or Docker image with reproducible solution, use of wandb to track experiments

Creative solutions and deployment (15 points)

• Creative Solutions (10 points): Novel approaches to food volume estimation (e.g. augmentation of the dataset with diffusion models).
• Deployment (5 points): creation of a deployed version of the app in a Gradio endpoint on HuggingFace

Resources Provided

Development Platform

• Google Colab: Pre-configured notebooks with quickstart tutorials

Technical Support

• Mentors with expertise in computer vision and food analysis
• Access to pre-trained food detection models and baseline implementations
• Documentation on food waste estimation techniques

Schedule

• 10:00 AM - Registration and team formation
• 10:30 AM - Technical briefing: Food waste estimation challenges
• 11:00 AM - Hacking begins
• 1:00 PM - Lunch break (observe your own food waste!)
• 2:00 PM - Mid-event check-in and mentor consultations
• 6:00 PM - Project submissions deadline
• 6:30 PM - Team presentations: Demo your food waste estimation system
• 7:30 PM - Judging and awards ceremony
• 8:00 PM - Event conclusion

Impact
Your work will contribute to understanding and reducing food waste through technology. The best solutions could help restaurants, cafeterias, and households track and minimize their environmental impact.

What to Bring

• Laptop with VSCode installed
• Basic knowledge of Python and machine learning frameworks
• Interest in sustainability and environmental impact

Join us to combine computer vision expertise with environmental consciousness in the heart of Potsdam's tech ecosystem!

Data Protection Notice
Please be aware that we will take photos and videos during the event. If you wish to opt out of this, please let us know in advance.