Welcome to the Visual Intelligence Laboratory (VI-Lab) Webpage. My name is Thanapong Intharah. I am an
assistant professor at the Department of Statistics, Khon Kaen University
(KKU).
I am the director of VI-Lab which works on applying Machine Learning and Computer Vision to
solve real world problems. Many of our projects were spun off as commercialized products: AI for CCTV and AI for Health .
I did my PhD at UCL
under supervision of Professor Gabriel
Brostow.
At UCL, I had been working on using Computer Vision to improve Programming by Demonstration, Learning to Automate GUI Tasks from
Demonstration
(project page).
Before joining KKU, I have founded a Deep Tech startup, TinyEpicBrains, where we work on invisible QR code. Currently, I am interested in Computer Vision for 3D data, video data and applications of AI in Healthcare.
I am also working on AI enhanced hardwares such as a differential dynamic microscopy and smart-glasses.
Visual domains such as Medical Imaging and Video Analysis are our expertise, but we also worked on
Chatbot, NLP, and other Machine Learning domains.
The following are list of selected research projects.
The OV-RDT platform consists of a mobile app, AI agents, and a dashboard.
The mobile app captures and detects OV-RDT test kits via smartphone cameras. AI ensures image quality and identifies Opisthorchiasis presence and severity.
Data, including AI diagnostics, are stored and displayed on a dashboard for analysis in epidemiology and patient screening progress.
Utilizing deep learning techniques, our project enhances dental care planning by accurately estimating age and gender from panoramic radiographs. Achieving 87.38% gender prediction accuracy and 1.96 years' age estimation error, it offers early diagnosis and forensic applications while ensuring model interpretability.
Valolyze: A Valorant gameplay analytics. Valolyze enables comprehensive analysis of Valorant gameplay from video recordings, empowering players and researchers alike in their quest for mastery and insights.
The KKU Herbarium Metaverse is a virtual simulation of the physical herbarium, offering a way to disseminate knowledge about plant diversity to a wider audience through an interactive platform that can display unlimited exhibition spaces and models of flowers regardless of the season, using photogrammetry techniques to create 3D models.
The development of a machine learning model for estimating the specific surface area (BET surface area) of carbon from plants using Scanning Electron Microscopy (SEM) images through deep learning is now available as a web application for predicting the BET surface area value, aiding in the use of renewable energy storage devices.
A system for determining human age from panoramic radiographs has been developed using deep learning techniques, resulting in a web application that is faster, more accurate, and reduces dental errors in diagnostic planning, dental and orthodontic treatment, and forensic identification.
We developed a web application that converts 3D models into voxel models with texture colors based on the color of the input models, for use in platforms such as game assets and non-fungible tokens (NFTs), providing the benefit of having voxel models with the same texture color as the input model.
In this work, we propose
dual image QR codes that aim to improve QR code capacity and appearance while
preserving the ability to be able to scan by standard QR code
readers.
In this work, we developed an end-to-end car damage estimation system via mobile camera.
To estimate the damage, we first build 3D models of a car with images from a mobile camera and compared the model through CNN.
Tagging questions according to their topics is useful
for internet forum management. In this work, we use the Bidirectional
Encoder Representations from Transformers (BERT)
model to categorize posts from Thai legal internet forums.
We developed a dataset by scraping from YouTube and manually annotated them. We then train an AI to
detect persons who concealed when they entered the scene.