Naveen Srinivasan

Clustered Image Data Pool using K-Means on their Feature characteristics to recommend more generic Training Data, increasing Training Data Quality and reducing Data preparation time by less than 25 percent. Extracted Image features using Mobilenet backbone and computed the Euclidean distance to retrieve similar Images. Implemented End to End pipeline of Sewer Tunnel Defect Discovery and Defect Validation Project using Tensorflow Object Detection API (EfficientDet) and deployed to Microsoft Azure Platform. Formulated an algorithm to Detect changes in Object state (Detect Incidents) by computing the Image Delta Histogram, enabling higher FPS and precision while reducing compute cost by less than 20 percent. Optimized the Image Sampling approach by filtering and processing images with probability of Object state change. Categorized Data Drift/Anomalies using Siamese Network Architecture. Experimented on Incremental Learning approaches. Leveraged knowledge in Data Preparation, Data Preprocessing, Hyper Parameter Tuning, and Training of Object Detection Models. Designed, Implemented, and Published Visualisations for presenting Dashboards. Built Backend wrappers for Custom Transformation of Data for Visualization. Guided the Annotation Team with suggestions for better quality of the Training Data.

Coursework: Machine Learning, Deep Learning, NLP, Big Data, Data Visualization, Statistics.