About
Dang-Khanh Nguyen
I am a developer and a learner. Check my coding activities at
.
My research record can be found at
or
.
Education
- 2022-2023
- Master Student, Department of AI Convergence; Chonnam National University (Gwangju, South Korea)
- Graduatded in Feb 2024.
- 2014-2018
- Bachelor of Engineering, Electrical and Electronic Engineering, Ho Chi Minh University of Technology (HCMC, Vietnam)
- Honor Program of Telecommunication Engineering.
Experience
- Senior Software Engineer at Aim Futrure
- 2024 - Present
- Design and implement SDK compiling well-known computer vision models (from various frameworks: Keras, Tensorflow, Torch) to hardware-specific runtime instructions.
- Simulate and verify hardware-specific executable file generated from pre-trained high-level models on dedicated simulator.
- Investigate lately released papers regarding computer vision models to deploy them on specific hardware platform.
- Researcher at Smart Computing Lab
- 2022 - 2023
- My research topic is video emotion recognition, multimodal learning, and multimodal fusion.
- Investigate and implement machine learning, deep learning model using Pytorch framework.
- My publications and competitions that I joined are listed in the next section.
- Software Engineer at Viettel High Technology Company
- 2020 - 2021
- Investigate RFC documents to implement network protocol. Develop L2/L3 Protocol for Network Device (Switch/Router) using C programming language.
- Some protocols that I have developed and tested: OMCI, SMNP.
- Use GDB for debugging and git to control version of team’s source code.
- Utilize Multi-thread, Inter-process communication to improve response time between devices.
- Simulate behavior of Intermediary network device for boundary-value analysis and stress test of Gateway device.
- Setup network topology to evaluate the behavior of the protocols on prototype device.
- Software Engineer at Renesas Design Vietnam
- 2018 - 2019
- Develop IP modules for virtual MCU/SoC using C++ programming language and SystemC.
- Some modules that I have developed and tested: Interrupt controller, Random Number Generator.
- Embed Python APIs in C++ module. Use Python to create test script and unit test environment.
- Guarantee code coverage >95% using GCOV. Track and resolve 100% leaking memory issues with Valgrind.
- Work in Linux Environment, use Makefile to control building process.
Publications
- Affective Behavior Analysis using Action Unit Relation Graph and Multi-task Cross Attention
- In Proceeding of ECCV Workshop (2022)
- The paper introduces my solution in the 4th Workshop and Competition on Affective Behavior Analysis in-the-wild. We achieve top 4 in the Multi-task Learning Challange.
- We propose a 3-head EfficientNet to resolve 3 affective tasks: emotion recognition, valence-arousal estimation, and action unit detection.
- Source code here.
- Fine-tuning Wav2vec for Vocal-burst Emotion Recognition
- In Proceeding of ACII Affective Vocal Bursts Workshop and Competition 2022 (A-VB)
- Analyzing Context and Speaker Memory using Pretrained Language Model for Emotion Recognition in Korean Conversation task
- In Proceeding of 10th International Conference on Big Data Applications and Services
- We achieve 4th Prize in the 4th Emotion Recognition in Korean Conversation Competition.
- We use pre-trained language model to analyze the conversation context and the speakers’ memory to handle the ERC task.
- Source code here.
- Multimodal Transformer for Automatic Depression Estimation System
- In Proceeding of The 29th International Workshop on Frontiers of Computer Vision
- We reviewed the transformer-based fusion methods of Xu for Depression recognition.
- We introduce a new fusion transformer and get good performance on D-Vlog Benchmark.
- Source code here.
- A Transformer-based Approach to Video Frame-level Prediction in Affective Behavior Analysis In-the-wild
- In Proceeding of 11th International Conference on Big Data Applications and Services
- The paper introduces my solution in the CVPR 2023: 5th Workshop and Competition on Affective Behavior Analysis in-the-wild. We achieved top 9 in the Multi-task Learning Challange.
- We used a pre-trained EfficientNet to extract facial spatial features and a Transformer to extract temporal features. The sequence of embeddings are then used to generate frame-wise emotional predictions for a video.
- Source code here.