Skip to content
Back to portfolio

Computer Vision

Brand Compliance Detection

A computer-vision pipeline that detects brand-compliance violations in video using frame-sequence analysis.

Tech Intern at Eli Lilly and Company · May 2025 – August 2025

Public descriptions omit internal video content, brand rules, proprietary evaluation data, and deployment details.

Problem

Brand-compliance violations in video needed to be detected using frame-sequence analysis.

My contribution

Developed a computer-vision pipeline using frame-sequence analysis to detect brand-compliance violations, evaluating the model with metrics appropriate for class imbalance and selecting PR-AUC over ROC-AUC because violation frequency was imbalanced.

Simplified architecture

Simplified: a frame-sequence computer-vision pipeline analyzes video for brand-compliance violations, evaluated with class-imbalance-aware metrics.

Outcomes

  • Evaluated on 8,400 frames across 210 videos with an 80/20 split.
  • 67 percent majority-class baseline.
  • 92 percent accuracy.
  • Precision of 0.89.
  • Recall of 0.87.
  • PR-AUC of 0.91.

Technologies

PythonOpenCVNumPyComputer vision