Overview
The context
Identifying ornaments in large volumes of images by hand is slow and error-prone. The pre-trained YOLOv8 model has no ornament classes, so an automated solution needed a purpose-built detector.
Challenge
What was holding them back
No ready-made model
Pre-trained detectors have no classes for ornaments.
Manual tagging
Reviewing and labelling images by hand took significant effort.
Fine visual detail
Ornaments are small, reflective and highly varied.
Objective
What they set out to achieve
- Detect ornaments reliably in input images
- Classify them accurately
- Process large image volumes automatically
- Reduce manual review effort and cost
Solution
What we built
Custom YOLOv8 training
We trained YOLOv8 on labelled ornament images to detect them in any input image.
ResNet-50 classification
A 50-layer residual network, pre-trained on ImageNet, refines classification.
Automated pipeline
Images flow through detection and classification without manual steps.
Scalable processing
Built to handle large image volumes efficiently.
Impact
The results
FasterOrnament identification — far less manual effort
AccurateDetection with fewer errors
ScalableHandles large image volumes
LowerLong-term labour cost
- Efficiency: automated detection replaces slow manual review
- Accuracy: image processing plus ML minimises identification errors
- Scalability and cost-effectiveness for growing catalogues