Jiawei Su

Senior Machine Learning Scientist

Fujitsu • Período integral

Aug 2020 - Mar 2022 • 1 yr 7 mos

Project: Construction and Implementation of Object Detection Models (Aug 2020 - Mar 2022) Role: Research Scientist (Team of approx. 50) Development Environment: Python, Linux, PyTorch, TensorFlow, Docker, GitLab, etc. Project Details: Object detection, classification, and segmentation in images/videos based on deep learning. [Project Overview] Recognizing phenomena centered on human and object actions from images/videos; proposing, building, and commercializing recognition models for retail, manufacturing, logistics, sports, etc. Examples: 1. Technology to sense human actions/expressions in videos to prevent marketing issues or accidents; technology to estimate attributes like age, gender, height. 2. Technology sensing human-object-environment relationships (e.g., "person talking to person," "person holding object"); inferring the meaning of surrounding objects (shelves, floors) from human actions in stores. Reference: https://www.fujitsu.com/jp/about/research/technology/actlyzer/achievement/ [Phases] Literature/tech research, proposal, requirements definition, basic design, R&D, etc. [Responsibilities] Literature and technology research and verification. Overall coding. Building the codebase in close collaboration with team members. PoC (Proof of Concept). [Achievements / Initiatives] 1. Built codebase, improved performance, and applied to business for Fujitsu's proprietary deep learning framework "Actlyzer." 2. Developed and commercialized an abnormal object detection and classification system for supermarkets and factories using models like R-CNN, YOLO. 3. Customized and improved models specialized in video object detection (RDN, MEGA, etc.) to enhance the performance of system 2). 4. Conducted tech research and created reports on Human-object interaction detection.