Half Hour of Flower: Planting the seeds for great Image Quality
summary
Royal FloraHolland builds a fully automated self‑improving image‑quality pipeline that uses VGG16 transfer learning for per‑property classifiers, wraps models in Docker APIs on AWS, stores artifacts in Artifactory and images on S3, and wires a feedback API VOC plus Airflow retraining. Decision logic called Judy combines model scores with product metadata. Performance tuning moves from 1600 ms to under 50 ms by using GPU instances (p3) and instance selection. Outcome: automated checks, monitored accuracy and an active feedback loop.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 4:37Dutch auction system
- 7:04Image quality check system
- 11:41Decision engine Judy
- 14:52Feedback loop with VOC
- 18:04Optimizing model performance
- 24:05Key learnings
- 26:57Q&A
related talks
17mEfficient & Frugal Image Data Processing Pipelines
01h 04mBuilding Efficient and Frugal Image Data Processing Pipelines
40mVertex AI: Pipelines for your MLOps workflows
20mQuality Control to avoid GIGO in Deep Learning Models
01h 00mOpening Keynote
21mTransform your web apps with Machine Learning
36mAI-Powered Real-Time Quality Control
33mHow to Track your Development Process with DVC