Hi there!

Featured works

• Built INT8 TensorRT engines for RF-DETR detection, 1.10–1.33× faster than FP16(opens in a new tab)

  roboflow/rf-detr9,791 stars

  quantization="int8" was silently ignored for TensorRT, so INT8 looked 13% slower than FP16. I traced the slowdown to Q/DQ placement and two TensorRT bugs, then wrote the placement myself on the FP16 graph. INT8 is now faster than FP16 at every size on an RTX 5070 (1.10–1.33×) and a Tesla T4 (1.11–1.21×), for 0.5–0.85 COCO AP points.

• Extended INT8 TensorRT to instance segmentation and keypoint models(opens in a new tab)

  roboflow/rf-detr9,791 stars

  INT8 export covered detection only, and the INT8 docs figures came from harnesses outside the repo. This PR accepts the five models and keeps the mask and keypoint heads in FP16. On an RTX 5070 at batch 1, INT8 is 1.06× (Seg Nano), 1.22× (Seg Small), 1.17× (Seg Medium), 1.09× (Seg Large) and 1.16× (Keypoint Preview) faster than FP16, for a mask AP loss of 0.47–0.91 points (keypoint AP loss 0.67 for Keypoint Preview). The docs tables are regenerated from a Colab-reproducible Jupyter cookbook that measures AP, latency and engine size on val2017.

• 

  roboflow/rf-detr

  RF-DETR is a real-time object detection and segmentation model architecture developed by Roboflow, SOTA on COCO, designed for fine-tuning. [ICLR 2026]

  Python9.8K26 PRs+24,652−3,132
• 

  roboflow/trackers

  Trackers gives you clean, modular re-implementations of leading multi-object tracking algorithms released under the permissive Apache 2.0 license. You combine them with any detection model you already use.

  Python3.9K2 PRs+239−8
• 

  sunsmarterjie/yolov12

  [NeurIPS 2025] YOLOv12: Attention-Centric Real-Time Object Detectors

  Python3K1 PR+1−1

Education

• 

  VNU-HCM University of Science

  2020 — 2024

  Computer Science, Honors Program.

  Focus: computer vision & vision–language models.

  • GPA: 8.67/10.0

  • Graduation Thesis: 10/10

    Developed an advanced Vision-Language system for simultaneous medical image segmentation and classification. By integrating visual and linguistic data, this research significantly enhances diagnostic accuracy and outperforms traditional anomaly detection models.

  • CV4DC Workshop at ACCV 2024

    Published "Abnormality Detection In Medical Image Based On Visual-Language Model" | Code.

    

    

    Previous slideNext slide

  • AI Challenge 2023 (HCMC)

    Award: Consolation prize | Code.

    

  • Ranked 6th in the 2020 University Entrance Exam

    Achieved a score of 28.65/30.00 | Link.

  2 more achievements
• 

  Phan Chau Trinh High School, Da Nang

  2017 — 2020

Stack

• TypeScript
• PostgreSQL
• JavaScript
• Python
• Go
• FastAPI
• MongoDB
• Redis
• Tailwind CSS
• SeaweedFS
• Docker
• Minio
• Prisma
• Express.js
• Nest.js
• React.js
• Next.js
• Kafka