• Established end-to-end automation test pipelines for evaluating computer vision models, from capturing data, executing model, running evaluation, gathering failure analysis breakdown, and generating visualizations; optimized for the large-scale dataset.
• Designed test cases and built regression datasets for sanity tests, which could quickly generate initial testing reports or find regression fields, to help shorten the test turnaround time.
• Developed tools to visualize computer vision models results, which give more intuitive examples on delivery reports to algorithm developers and management, to help understand the models' performance and triage the failure pattern.
• Developed HTML-based test report delivering the overall model performance and detailed breakdown metrics to help the audience directly get information.
• Built PostgreSQL database + web-based visualization querying tools, which store evaluation metrics, to show the clear trend of model performances
