Skip to content
View ShuaiLYU's full-sized avatar
🏠
Working from home
🏠
Working from home
74 followers · 9 following
  • Ultralytics
  • Hong Kong
  • 11:40 (UTC -12:00)

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull SharkAchievement: Starstruckx3Achievement: Arctic Code Vault Contributor

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull SharkAchievement: Starstruckx3Achievement: Arctic Code Vault Contributor

Highlights

  • Pro

Block or report ShuaiLYU

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ShuaiLYU/README.md

Hi there, I'm Shuai LYU (Louis) 👋

I'm a computer vision researcher and engineer focused on industrial defect detection, anomaly detection, and few-shot learning. I'm actively contributing to Ultralytics, working on open-vocabulary detection (YOLOE).

� Experience & Education

Ultralytics · Full-time · Shenzhen, Guangdong, China

  🔹 Senior Machine Learning Engineer · Jan 2026 – Present

  🔹 Machine Learning Engineer · Aug 2025 – Jan 2026 · 6 mos

🎓 The Hong Kong Polytechnic University

  🔹 Doctor of Philosophy (PhD) · Sep 2021 – Jun 2025 · 3 yrs 9 mos

🎓 Guangdong University of Technology

  🔹 Master of Engineering · Jun 2017 – Jun 2020 · 3 yrs

🔭 Work Direction @ Ultralytics

I contribute to the ultralytics/ultralytics project, mainly in the following areas:

  • 🏷️ YOLOE — Open-vocabulary / prompt-free object detection: visual prompt training, text model support, multi-config YAML training, memory bank for inference
  • 🎯 SAM2 — Interactive segmentation: SAM2DynamicInteractivePredictor for few-shot interactive inference
PR Title Status
#23592 Improve Results.save() with pathlib and optional directory creation ✅ Merged
#23427 Support multiple data configs via YAML for YOLOE training ✅ Merged
#23428 Fix YOLOE text model attribute in trainer from scratch ✅ Merged
#23401 Fix visual prompt training for YOLOE26 ✅ Merged
#23046 Fix loss name box_loss → cls_loss in TVPDetectLoss ✅ Merged
#21947 Split large channel masks to handle cv2.resize 512 limit ✅ Merged
#21745 Fix YOLOE prompt-free validation example in docs ✅ Merged
#21232 SAM2: Add SAM2DynamicInteractivePredictor for few-shot inference ✅ Merged
#22255 Add memory bank for YOLOE predict 🔄 Open

📄 Papers

Google Scholar

  • MVREC: A General Few-shot Defect Classification Model Using Multi-View Region-Context AAAI 2025 · arXiv · Code A CLIP-based few-shot framework for multiclass industrial defect classification, introducing the MVTec-FS benchmark (1228 images, 46 defect types).

  • REB: Reducing Biases in Representation for Industrial Anomaly Detection Code Self-supervised representation learning with DefectMaker synthetic augmentation and LDKNN for unsupervised anomaly detection on MVTec AD & MVTec LOCO.

�🛠️ Skills & Tools

Python PyTorch OpenCV

Research Areas: Industrial Defect Detection · Anomaly Detection · Few-Shot Learning · Open-Vocabulary Detection · Interactive Segmentation

📫 Contact

Pinned Loading

  1. ultralytics/ultralytics ultralytics/ultralytics Public

    Ultralytics YOLO 🚀

    Python 54.4k 10.5k

  2. MVREC MVREC Public

    Official source code for MVREC: A General Few-shot Defect Classification Model Using Multi-View Region-Context.(AAAI 2025)

    Python 37 2

  3. REB REB Public

    REB：Reducing Biases in Representation for Industrial Anomaly Detection

    Jupyter Notebook 26 6

  4. Deep-Learning-Approach-for-Surface-Defect-Detection Deep-Learning-Approach-for-Surface-Defect-Detection Public

    A Tensorflow implementation of "Segmentation-Based Deep-Learning Approach for Surface-Defect Detection"

    Python 752 261