VQGAN-based Automatic Scoring for Beery-VMI

Background & context

The Beery-VMI is a clinical test of visual-motor integration, scored by hand from a person's drawings — slow, subjective, and with scarce training data for this population.

This project built an AI system to score the test automatically, using VQGAN to synthesize additional training images and solve the data-scarcity problem.

Key responsibilities & role

• Data pipelineBuilt the data pipeline for Beery-VMI scoring.
• Synthetic dataUsed VQGAN / GAN to generate synthetic test images, expanding a small dataset.
• ModelingTrained the scoring model on the augmented dataset with PyTorch and OpenCV.
• ValidationChecked score consistency against manual grading.

Tools & knowledge

Results & achievements

→ Presented at a 2022 academic conference.

→ Published at ICGHIT 2023: “Development of an Automatic Scoring System for Beery-VMI based on Artificial Intelligence.”