Restaurant Kitchen Insect Detection
A Synthetic-Data Pipeline For Training A YOLO Insect Detector.
The Problem
Training A YOLO-Based Insect Detector For CCTV Footage In Restaurant Kitchens Needs Labeled Data, But No Suitable Public Dataset Exists For That Environment. Hand-Collecting And Annotating Thousands Of Real Images Of Flies, Cockroaches And Ants In Kitchens Was Impractical.
What We Built
Built A Python-Based Compositing Pipeline That Automatically Overlays Insect Cutouts (Flies, Cockroaches, Ants) Onto Real Kitchen Backgrounds With Randomized Scale, Rotation And Position.
CCTV-Style Effects, Noise, Blur And Desaturation, Are Applied To Match Real Deployment Conditions, So The Model Trains On Images That Look Like The Camera Feed It Will Actually Run On.
The Pipeline Auto-Generates YOLO-Format Bounding-Box Annotations For Every Image, Producing 500+ Labeled Training Images In Minutes, And Splits The Dataset Into Train/Val/Test Sets With A Ready-To-Use data.yaml For Direct YOLO Training.
Key Highlights
- ✓Solved A No-Public-Dataset Problem With Synthetic Data
- ✓500+ Labeled Images Generated In Minutes
- ✓Auto-Generated YOLO-Format Bounding-Box Annotations
- ✓Realistic CCTV Effects: Noise, Blur, Desaturation
- ✓Train/Val/Test Split With Ready-To-Use data.yaml
