Heng Zhang

PhD student

heng.zhang -AT- inria.fr

Paris, France

About Me

I am a final year PhD student in IRISA/INRIA Rennes laboratory and University of Rennes 1. I am working under the supervision of Prof. Elisa FROMONT and Prof. Sébastien LEFEVRE. In the same time, I work as a Deep learning R&D Engineer at ATERMES in Paris. My current research interest is deep learning for object detection, multi-sensor fusion, active learning and knowledge distillation.

Work Experience

• Deep Learning R&D Engineer

  ATERMES | Paris, France | Dec. 2018 - ??

  ▮ Built Deep Learning models for accurate object detection (car, pedestrian, etc) at long distance (>1.5km).

  ▮ Adaptively fusing information from multiple sensors (e.g., thermal camera & visible camera) to improve the detection precision.

  ▮ Accelerate CNN model's inference speed for efficient DL integration under embedded constraints.

• Training contract (one year)

  Hubert Curien laboratory | Saint-Etienne, France | Sep. 2017 - Oct. 2018

  ▮ Implement different deep learning models (Faster RCNN, SSD, YOLO, RetinaNet) for face/person detection in public transport.

  ▮ Proposed efficient video object detection methods for video surveillance applications.

Education

• Ph.D. student in Deep Learning and Computer Vision

  University of Rennes 1 | Rennes, France | Dec. 2018 - ??

  ▮ Subject: Multispectral object detection.

  ▮ Industrial research training program, in cooperation with ATERMES company.

• Engineer's degree (equivalent to a master's degree)

  Télécom Saint-Etienne | Saint-Etienne, France | Sep. 2015 - Oct. 2018

  ▮ Major: Computer Science and Computer Vision

  ▮ One-year professional training contract with Hubert Curien laboratory.

• Bachelor's degree

  Xidian University | Xi'an, China | Sep. 2012 - Jun. 2016

  ▮ Major: Electronic Engineering

Publications

• "Low-cost Multispectral Scene Analysis with Modality Distillation", in WACV 2022

  Heng Zhang, Elisa FROMONT, Sébastien LEFEVRE, Bruno AVIGNON

  ▮ Keywords: Multispectral Scene Analysis, Thermal Sensor, Knwoledge Distillation

  [PDF] [Poster]

• "PDF-Distil: Including Prediction Disagreements in Feature-based Knowledge Distillation for Object Detection", in BMVC 2021

  Heng Zhang, Elisa FROMONT, Sébastien LEFEVRE, Bruno AVIGNON

  ▮ Keywords: Object Detection, Knwoledge Distillation, Model Compression

  [PDF] [Code] [Slides]

• "Deep Active Learning from Multispectral Data Through Cross-Modality Prediction Inconsistency", in ICIP 2021

  Heng Zhang, Elisa FROMONT, Sébastien LEFEVRE, Bruno AVIGNON

  ▮ Keywords: Multispectral Scene Analysis, Active Learning

  [PDF] [Video]

• "Guided Attentive Feature Fusion for Multispectral Pedestrian Detection", in WACV 2021

  Heng Zhang, Elisa FROMONT, Sébastien LEFEVRE, Bruno AVIGNON

  ▮ Keywords: Multispectral Object Detection, Attentive Fusion, Multi-sensor Fusion

  [PDF] [Results] [Video] [Demo]

• "Localize to Classify and Classify to Localize: Mutual Guidance in Object Detection", in ACCV 2020

  Heng Zhang, Elisa FROMONT, Sébastien LEFEVRE, Bruno AVIGNON

  ▮ Keywords: Object Detection, Label Assignment, Anchor Matching

  [PDF] [Code] [Video] [Demo1] [Demo2]

• "Multispectral Fusion for object detection with Cyclic Fuse-and-Refine blocks", in ICIP 2020

  Heng Zhang, Elisa FROMONT, Sébastien LEFEVRE, Bruno AVIGNON

  ▮ Keywords: Multispectral Object Detection, Cyclic Fusion, Multi-sensor Fusion

  [PDF] [Results] [Video] [Demo]

• "Méthodes de Fusion pour la Détection Multispectrale par Réseaux de Neurones Profonds", in RFIAP 2020

  Heng Zhang, Elisa FROMONT, Sébastien LEFEVRE, Bruno AVIGNON

  ▮ Keywords: Multispectral Object Detection, Multi-sensor Fusion

  [PDF] [Video]

• "Improving video object detection by Seq-Bbox Matching", in VISAPP 2018

  Hatem Belhassen, Heng Zhang, Virginie Fresse, El-Bay Bourennane

  ▮ Keywords: Video Object Detection, Bounding Box Association

  [PDF] [Demo]