Youtube VIS Video Instance Segmentation 2021 Challenge

Conducted at CyberCore when I was a Machine Learning Engineer

  • Challenge description
  • Time: Feb. 2020 – now
  • Role: Team member of a team with 3 members.
  • Description: I mainly focus on reading and implementing papers to improve the model results on three tasks: Object Detection, Instance Segmentation, and Object Tracking.
  • Result: In process

Transformer-based Multiple Object Tracking

Conducted at CyberCore when I was a Machine Learning Engineer

  • Time: Sep. 2020 – now
  • Role: Team member of a team with 3 members.
  • Description: Research on Transformer (Attention Is All You Need) methods Multiple Object Tracking problem, benchmarked on Autonomous Driving datasets.
  • Result: In process

Transformer-based for Seeing Without Looking

Conducted at CyberCore when I was a Machine Learning Engineer

  • Time: Jan. 2021 – Mar. 2021
  • Role: Main contributor of a team of 2 members.
  • Description: Improve the model performance by replacing the RNN network in the original Seeing Without Looking papers.
  • Result: Reach small improvements.

Line Segment Detection

Conducted at CyberCore when I was a Machine Learning Engineer

  • Time: Dec. 2020 – now
  • Role: Team member of a team with 3 members leading by Thuy Nguyen Chinh
  • Description: This is a project between CyberCore and TODA. In this project, I mainly focus on developing a fast and robust model processing Full-HD images on CPU with 3~10 fps for checking steel structure in construction.
  • Result: In process

Zalo AI Challenge 2020 - Traffic Sign Detection

Conducted at CyberCore when I was a Machine learning Engineer

  • Challenge description
  • Time: Middle of Nov. 2020 – The end of Nov. 2020 (2 weeks)
  • Role: Corresponding member with 3 other members Thuy Nguyen Chinh, Son Nguyen Thanh, and Chuong Nguyen Hoang.
  • Description: This is a yearly online competition for Vietnam’s AI engineers to explore AI technologies and impact life in exciting new ways hold by Zalo. Our team name: CyberCoreVN
  • Result: Rank 1st in Public Leaderboard, Rank 3rd in Private Leaderboard

Multiple Object Tracking

Conducted at CyberCore when I was a Machine learning Engineer

  • Time: Aug. 2020 – Now
  • Role: Team member of a team with 3 members leading by Thuy Nguyen Chinh and Chuong Nguyen
  • Description: This is an internal research project of the company. The datasets we need to make benchmark are Waymo, BDD100k, and MOT16/17. The target conferences are CVPR, ICCV, ECCV, or NeurIPS.
  • Result: On going

Rebar-Knot semantic segmentation

Conducted at CyberCore when I was a Machine learning Engineer

  • Time: Jul. 2020 – Aug. 2020
  • Role: Team member of a team with 2 members leading by Thuy Nguyen Chinh
  • Description: This is a project between CyberCore and TODA. In this project, I mainly focus on finding out a small model with acceptable accuracy and run at least 10 fps on CPU i5-9400; implementing various loss functions and adjust the dataset’s labels.
  • Result: The model is delivered to the customer.

3D Dangerous Object Detection using Milliwave Radar

Conducted at CyberCore when I was a Machine learning Engineer

  • Time: Jun 2020 – now
  • Role: Team member of a team with 6 members leading by Thuy Nguyen Chinh and Tuan Ho
  • Description: Cooperate with Taiyo Yuden to develop a security product for early alert at airports. It uses a network of various milliwave radars to detect dangerous objects (e.g., knife, gun) inside clothes.
  • Result: In progress.

Head Pose Estimation

https://github.com/TuanTNG/Cybercore-Head-Pose-Estimation

Conducted at CyberCore when I was a Machine learning Engineer

  • Time: May 2020 – Jul. 2020
  • Role: Major contributor of a team of 2 members, led by Mr. Khai.
  • Description: We researched, read papers, implementing network architectures and modify the training dataset to help the network learning easier as well as more stable. This project is a demo of CyberCore delivered to customer in Japan.
  • Result: The demo video was sent to the customer.

Vehicles Detection, Tracking and Speed Estimation

https://github.com/TuanTNG/VDTSE

Conducted at Ho Chi Minh City University of Technology when I was a student

  • Time: Dec. 2019 – Sep. 2020
  • Role: Main contributor of a team of 2 members.
  • Description: This is my Thesis. Our team researched and applied SOTA algorithms for object detection, tracking and speed estimation of vehicles at the overpass in Ho Chi Minh City, Vietnam. The dataset is collected and labeled by team members.
  • Result: The highest grade in the Council.

Implement YOLOv3 from scratch

Conducted at Ho Chi Minh City University of Technology when I was a student

  • Time: Dec. 2019 – Feb. 2020
  • Role: Research.
  • Description: Reading YOLOv1/2/3 papers, implementing Neck, Head and Loss Functions of YOLOv3 from scratch using Pytorch.
  • Result: Stop training at epoch 17 with 0.16 mAP because of limited GPU resources. For class person only, the model achieved 0.42 in AP.