Industrial Anomaly Detection & Vision Encoder

Led large-scale (1M+ data) vision encoder pre-training to optimize feature extraction, improving Anomaly Detection AUC by 15% for tasks with extreme data imbalance.

Anomaly DetectionComputer VisionRepresentation LearningImbalanced Learning

Industrial Anomaly Detection

At PEGAAi, I addressed critical quality control challenges in manufacturing by developing robust vision-based anomaly detection systems.

Key Contributions

  • Vision Encoder Optimization: Led large-scale (1M+ data) vision encoder pre-training and fine-tuning to optimize feature extraction tailored to industrial defects.
  • Extreme Data Imbalance: Designed solutions for scenarios with highly skewed data distributions and limited NG (defective) samples.
  • Performance Improvement: Successfully improved Anomaly Detection AUC by 15% across challenging manufacturing datasets.
← 返回專案