Jieming Song

Developed a multi-camera customer tracking system on tensorflow based on RetinaNet and deep-sort in shopping mall Established CNN model to recognize pedestrian attributes and achieved customer lookup from photo or attributes Analyzed path and group behavior of each customer to predict his intention and updated prediction by online data Achieved multi-camera re-identification based on single-camera tracking and 2-d geometric mapping by camera calibration

Built deep-learning based parts auto grabbing system by localization and angle detection from object detection network Researched and trained Faster-rcnn network and revised layer structure to output part position, type and angle Completed a pipeline from data collecting and augmentation to network structure design, training, and optimization