Khanh is currently unavailable
Khanh Ha
Verified Expert in Engineering
Software Engineer and Developer
Hanoi, Vietnam
Toptal member since January 7, 2020
Bio
Khanh Ha is a software engineer veteran with a track record of industry-level projects in the fields of computer vision, deep learning, machine learning, 3D geometry, and computer graphics technologies. He also specializes in digital human technologies, from 3D avatar estimation to motion capture.
Portfolio
MoveAI
C++, Python, JavaScript, Blender Scripting API, Qt, PyTorch, NVIDIA TensorRT...
The Ohzone, Inc.
Blender, C++, OpenCV, PyTorch, Deep Learning, Python, Computer Vision
Computer Vision Lab, Bauhaus University
OpenCL, OpenCV, Keras, PyTorch, Deep Learning, Python, C++, Computer Vision
Experience
- C++ - 10 years
- OpenGL - 8 years
- Computer Graphics - 6 years
- Deep Learning - 5 years
- Qt - 5 years
- OpenCV - 5 years
- WebGL - 5 years
- Babylon.js - 2 years
Work Experience
Computer Vision Consultant
2019 - 2023
MoveAI
- Developed a real-time motion capture system that can support more than 10 cameras and stream data to Unity and Unreal Engine at 60 FPS.
- Pushed AI models beyond the normal boundaries to achieve precise 2D human key points from RGB human images under challenging edge cases such as markerless, strong occlusion, and low resolution.
- Created kinematic-based human motion tracking solutions for reconstructions of high-quality motions of multiple humans under close interaction and complex movement contexts.
- Developed performance-tracking solutions for sports that can track up to 10 players with an accuracy above 90%.
Technologies: C++, Python, JavaScript, Blender Scripting API, Qt, PyTorch, NVIDIA TensorRT, Deep Learning, QML, OpenCV, Computer Vision, Unreal Engine 4
Senior Computer Vision Engineer
2017 - 2019
The Ohzone, Inc.
- Built a robust deep learning model for image-based 3D human reconstruction that can tackle various real cases and achieve more accurate results than competitors.
- Developed image processing for a state-of-the-art 3D clothes reconstruction project that can reconstruct high-detail 3D clothes models.
- Created mesh processing algorithms to improve the quality of reconstructed meshes.
Technologies: Blender, C++, OpenCV, PyTorch, Deep Learning, Python, Computer Vision
Computer Vision Engineer
2018 - 2018
Computer Vision Lab, Bauhaus University
- Built a crack segmentation deep learning model that outperforms state-of-the-art models in detecting edge cases that cause a high rate of false-positive in reality.
- Developed a new optimized dense stereo matching pipeline to handle high-resolution stereo images.
- Created a stereo reconstruction rig from scratch that can construct a stereo real-time depth map at 60 FPS.
Technologies: OpenCL, OpenCV, Keras, PyTorch, Deep Learning, Python, C++, Computer Vision
Senior Mesh and Graphics Developer
2014 - 2016
Nyomo
- Built an OpenGL-based Graphics Engine for the 3D printer software Nyomo that improved customer experience through realistic mesh visualization methods.
- Developed mesh processing methods such as re-meshing, smoothing, and decimation that helped customers achieve better mesh topology without sacrificing the mesh shape.
- Created digital sculpting features for 3D printer software to give customers more flexibility in modifying their 3D meshes.
Technologies: Computer Graphics, OpenGL, Qt, C++, WebGL
Back-end Developer
2012 - 2013
Admicro
- Developed a logging back-end server based on C++ and Lua for an online advertising system that serves around 50K of requests per second and helps to integrate new business features seamlessly without downtime.
- Implemented business features based on Lua for an online advertisement system.
- Maintained and optimized back-end services for an online advertisement system.
Technologies: SQL, Redis, Back-end, Boost, C++
Project History
Real-time Motion Capture System
https://www.youtube.com/watch?v=E7HkaKMLDNU&list=PLkLcpRjubtMcwnF_R7rO1u5JFHsOkDP1RA human motion capture system that can run in real-time at 60 frames per second (FPS) with more than eight GigE cameras. The system can run distributed mode on multiple PCs to support more cameras at high FPS. The mocap data can also be previewed in-app or streamed to Unreal Engine for preview.
3D Printer Software
https://www.youtube.com/channel/UChObDqXzc4RD5gLZmF_Ey3wDuring my time at Nyomo, a 3D printer software company based in Singapore, I developed the graphics module for the software that visualizes the 3D objects to the user to make it easier for users to distinguish the small details and curvature of 3D object surfaces. The graphics module that I wrote is also used to visualize other software functionalities. Above is a quick tour of the main features of the software.
Crack Segmentation
https://github.com/khanhha/crack_segmentationIn this project, I developed a deep learning model for crack segmentation for a bridge-investigation project using drones. One challenge I solved in this project is the high false-positive rate due to crack-similar patterns in real-world images collected by drones at different lighting, zoom-in, and perspectives. The trained model can detect cracks with high precision and avoid similar patterns like planking lines commonly found on bridge surfaces.
For more information about the project, please visit the website: https://khanhha.github.io/portfolio/crack-segmentation/.
For more information about the project, please visit the website: https://khanhha.github.io/portfolio/crack-segmentation/.
Deep Learning Based 3D Human Avatar Estimation
https://github.com/khanhha/human_avatar_reconstructionA system for estimating 3D human avatars from photos taken by a mobile phone. The output includes a ready-to-use mesh that can be used for virtual try-on. The model also has a 3D face with the customer's facial texture.
Skeleton Meshing
https://developer.blender.org/D1465This is a feature that I built for Blender, the most popular 3D software. This feature is used by artists to create a rough mesh from a skeleton structure, which helps speeds up the modeling process and free artists from doing labor-intensive jobs.
Voxel Artwork Importer and Viewer
https://www.youtube.com/watch?v=0XX7ImySkNw&list=PLkLcpRjubtMfBnvqnge646SncXCaHWqBJ&index=1I developed a Blender add-on for importing .vox files from the voxel art builder software to Blender with optimized meshes, scene hierarchy, and similar material.
I created baking tools for transforming the imported voxel files in Blender to a lightweight, portable format that can be run on web browsers with true lighting information. Also, I built a WebGL app to display voxel artwork for the user to explore .vox files.
I created baking tools for transforming the imported voxel files in Blender to a lightweight, portable format that can be run on web browsers with true lighting information. Also, I built a WebGL app to display voxel artwork for the user to explore .vox files.
Education
2018 - 2020
Master's Degree in Computer Science
Bauhaus University Weimar - Weimar, Germany
2007 - 2012
Engineer's Degree in Computer Science
Hanoi University of Science and Technology - Hanoi, Vietnam
Certifications
AUGUST 2018 - PRESENT
Nanodegree in Deep Learning
Udacity
Skills
Libraries/APIs
PyTorch, TensorFlow, OpenCV, Scikit-learn, OpenGL, Keras, Node.js, Blender Scripting API, WebGL, Babylon.js, Three.js
Tools
Blender, MATLAB
Languages
C++, SQL, JavaScript, HTML, QML, Python
Paradigms
Object-oriented Programming (OOP), Functional Programming, Distributed Computing, Parallel Computing, Model View Controller (MVC)
Frameworks
Qt, Boost, OpenCL, Django, Unreal Engine 4, Unity
Platforms
Linux, Windows
Storage
Redis, MySQL, MongoDB
Other
Machine Learning, Unsupervised Learning, Clustering Algorithms, Clustering, Regression Modeling, 3D Pose Estimation, Probabilistic Graphical Models, Tracking Systems, Image Processing, Computer Graphics, Deep Learning, Back-end, NVIDIA TensorRT, Virtual Reality (VR), Computer Vision, Computer Science, Computational Geometry, 3D, Voxel
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring