CV

Contact Information

Name Junfeng Ren
Professional Title M.S. Student in Electronic Information
Email junfengren3253@gmail.com

Professional Summary

M.S. student at Southern University of Science and Technology (SUSTech), working on computer vision for autonomous driving. My research focuses on collaborative perception, 3D semantic occupancy prediction, and efficient multi-agent communication.

Experience

  • 2023 - 2024

    Qingdao, China

    Research Assistant
    IoT Engineering Laboratory, Shandong University of Science and Technology
    • Conducted research on real-time scheduling optimization for embedded IoT systems under multi-task environments
    • Improved the Rate Monotonic Scheduling (RMS) algorithm to enhance task scheduling efficiency and system real-time performance
    • Implemented scheduling algorithms and built simulation environments to evaluate performance under varying workloads and priority settings
    • Analyzed system performance in terms of response time and resource utilization, validating the effectiveness of the proposed method
    • Co-authored an IEEE paper: ‘Scheduling Optimization Design of IoT Embedded System Based on Improved RMS Algorithm’

Education

  • 2024 - Present

    Shenzhen, China

    M.S.
    Southern University of Science and Technology
    Electronic Information
    • Research focus: Computer Vision and Autonomous Driving Perception
    • Topics: Collaborative Occupancy Prediction, Multi-Agent Perception
  • 2019 - 2023

    Qingdao, China

    B.S.
    Shandong University of Science and Technology
    Internet of Things Engineering
    • Focus on embedded systems, robotics, and computer vision applications

Publications

  • 2026
    Learning to Merge Tokens for Communication-Efficient Collaborative Occupancy Prediction
    In preparation
  • 2026
    Rate-Distortion in Efficient Multi-Agent Perception: A Unified Framework for Communication and Memory Optimization
    In preparation
  • 2023
    Scheduling Optimization Design of IoT Embedded System Based on Improved RMS Algorithm
    IEEE

Skills

Programming (Advanced): Python, PyTorch, CUDA, Linux, C/C++
Machine Learning (Advanced): Deep Learning, Transformers, Reinforcement Learning
Computer Vision (Advanced): 3D Perception, Occupancy Prediction, Autonomous Driving, World Model

Languages

Chinese : Native speaker
English : Fluent

Interests

Research Interests: Computer Vision, Autonomous Driving Perception, 3D Scene Understanding, Semantic Occupancy Prediction, Collaborative Perception, Multi-Agent Systems

Certificates

  • IELTS - British Council (2024)
  • Software Copyright - IoT Data Monitoring and Analysis System - National Copyright Administration of China (2023)

Projects

  • Communication-Efficient Collaborative 3D Occupancy Prediction (LiteTokenOcc)

    Research on communication-efficient multi-agent 3D semantic occupancy prediction for autonomous driving, focusing on tokenized scene representation, spatio-temporal memory, and request-driven communication under bandwidth constraints.

    • Proposed a multi-agent collaborative 3D occupancy prediction framework based on tokenized scene representations
    • Designed a spatio-temporal memory module to model and reuse information across time and vehicles
    • Developed a request-driven communication mechanism for selective information exchange among agents
    • Introduced communication-aware token merging to compress bandwidth from relevance, reliability, and temporal novelty
    • Achieved strong perception performance while reducing communication cost to the KB level on Occ3D-nuScenes and Semantic-OPV2V
    • Extending occupancy representation toward world model learning for unified perception and temporal modeling
  • Rate-Distortion Optimization for Efficient Multi-Agent Perception

    Research on an information-theoretic framework for efficient multi-agent perception, modeling the trade-off between communication bandwidth and perception performance under constrained settings.

    • Formulated collaborative perception as a rate-distortion optimization problem under bandwidth constraints
    • Developed a unified framework to analyze trade-offs among token compression, information selection, and model performance
    • Designed information-aware compression mechanisms combining token pruning and spatio-temporal memory
    • Systematically evaluated performance under different communication budgets (KB level)
    • Reduced communication bandwidth by ~70% while keeping performance degradation within 10%
    • Provides theoretical foundation for communication-efficient perception and world model representation learning
  • Federated Learning for Medical Image Segmentation with Domain Generalization

    Research project on privacy-preserving medical image segmentation under cross-institution domain shift, combining federated learning and domain generalization techniques.

    • Implemented a federated learning framework (FedAvg) for multi-institution medical image segmentation without sharing raw data
    • Designed domain generalization strategies including style transfer augmentation and feature distribution alignment
    • Built a U-Net–based segmentation pipeline and evaluated cross-domain generalization performance
    • Analyzed the interaction between federated learning and domain shift across heterogeneous datasets
  • Autonomous Driving Perception and Decision System on NVIDIA Jetson

    Embedded autonomous driving system integrating perception, decision-making, and control on NVIDIA Jetson.

    • Built a perception pipeline using YOLOv5 for object detection and OpenCV-based lane detection (Canny + Hough)
    • Designed a rule-based decision module for driving behaviors such as lane following, turning, and obstacle avoidance
    • Implemented a finite-state machine (FSM) for stable and interpretable behavior transitions
    • Developed closed-loop control for steering and speed adjustment
    • Optimized model inference with TensorRT for real-time performance on embedded GPU
  • Embodied AI Soccer Task System on NAO Robot

    Vision-driven embodied AI system for autonomous ball detection, navigation, and kicking on a humanoid robot.

    • Implemented real-time ball detection using HSV segmentation and contour analysis
    • Estimated target distance using a pinhole camera model and geometric reasoning
    • Designed a finite-state machine (FSM) for search, approach, alignment, and kicking behaviors
    • Achieved closed-loop control with visual feedback
    • Used NAOqi APIs for motion control including walking, turning, and kicking
  • Embedded Edge AI Face Recognition Access Control System

    Edge AI system for real-time face recognition and access control on resource-constrained embedded devices.

    • Built a lightweight face recognition pipeline using MobileFaceNet and cosine similarity matching
    • Applied model pruning and INT8 quantization for TinyML deployment on STM32
    • Designed on-device inference and local decision-making for low-latency and privacy
    • Integrated ESP8266 for WiFi communication with backend services (MQTT/HTTP)
    • Developed an Android app for user management and remote control