Nyan Swan Aung
Projects
I have worked on some truly inspiring projects while networking with many influential and creative individuals along the way. I have completed on a number of personal projects, school projects, internship tasks and collaborative projects with around the world in the fields of web development, software and artificial intelligence. For more information about my career experience and current work portfolio, have a look at my past projects.
Product Lead | Collaborative Project | Machine Learning
March 2023 ~ May 2023
During this 8 week project, I had an opportunity to work with AI Engineers from around the world in, a Germany based precision vaccine company, Belyntic's project which is held by Omdena.
​
Belyntic creates novel self-adjuvant vaccines based on synthetic peptides in the space of viral diseases. The focus lies on diseases with a clear unmet medical need. The goal of the project was focused on developing a model for a therapeutic vaccine against a deadly disease without a cure, called Progressive multifocal leukoencephalopathy (PML).
​
As a product lead in an AI project, I played a crucial role in facilitating collaboration between machine learning engineers and data scientists. This involved working closely with the team to define product requirements, prioritize features, and ensure timely delivery. Additionally, I took the lead in overseeing a cross-functional team, driving the development and implementation of an AI analysis tool aimed at accelerating vaccine development. To ensure efficient task management, I utilized project management tools like Miro and Trello to assign tasks to team members with clear deadlines. Furthermore, I conducted weekly sprint meetings to foster effective communication and alignment within the team, ensuring everyone was on the same track towards project success.
Lead ML Engineer | Freelance Project | Computer Vision
July 2022 ~ August 2022
During this 2 months project, I had an opportunity to work on, a Mexican based company Solvento, AI project.
​
In the field of logistics, manually verifying signatures, stamps, and invoice IDs for a large number of deliveries can be a cumbersome and time-consuming process. The project aims to address this challenge by developing an automated system that leverages AI to detect and verify signatures, stamps, and invoice IDs on PDF documents. By automating this task, we aim to streamline the verification process, saving time and effort for logistics professionals and improving overall efficiency.
Lead ML Engineer | Collaborative Project | Computer Vision
June 2022 ~ August 2022
During this 8 week project, I had an opportunity to work with AI Engineers from around the world in, a UK tech-startup, Drone Ag's project which is held by Omdena.
​
DroneAg’s flagship product Skippy Scout is a mobile application that automates the flight of a drone to capture data that is used to provide analysis and insights to farmers and agronomists.
The goal of this project is to build a segmentation machine learning model in order to extract meaningful insights about crops, such as health, type, and area coverage.
​
In this project, I actively contributed to various aspects of the development and deployment of advanced computer vision models such as annotating mask images, data preprocessing, post-processing, dataset management, training and inferencing Unet, Unet++, Deeplabv3+ using AWS Sagemaker, Kaggle, and Colab, parameter tuning and benchmarking using the Weight&Bias platform. Finally, I successfully deployed an AI application using Docker
Lead ML Engineer | Collaborative Project | Computer Vision
Jan 2022 ~ March 2022
During this 8 week project, I had an opportunity to work with AI Engineers from around the world in, a Switzerland based tech-startup, Archilyse's project which is held by Omdena.
​
To recognize floor plan elements in a layout requires manual labor to draw the different elements over the image. The goal of this project has been to improve the efficiency of this manual effort by automatically identifying the relevant types of objects present using state-of-the-art deep learning and computer vision approaches.
​
In this project, I have successfully implemented the project pipeline by reading research papers, training/benchmarking YOLOX and YOLOV5 object detection models, dockerizing project pipeline, EDA and converting polygons to bbox COCO dataset format.
Collaborative Project | Computer Vision
10/12/2020 ~ 4/2/2021
During this 8 week project, I had an opportunity to work with AI Engineers from around the world in this, a tech-startup, WeedBot's project held by Omdena.
​
The goal is to develop a high-speed plant image recognition neural network with a speed of 12ms per image or faster and recognition precision of 100-110% of crop polygon (which means up to 10% false positives are allowed).
​
In this project, my tasks were building the weed segmentation model using semantic architectures and data annotations. Moreover, I got a chance to learn about COCO annotation, TensorRT conversion, training on AWS and benchmarking on NVIDIA Jetson hardware.
Collaborative Project | Computer Vision
18/3/2021 ~ 29/5/2021
During this 8 week project, I had an opportunity to work with data scientists from around the world in this project owned by precision livestock farming company called, Faromatics which is held by Omdena.
​
The goal is to implement an individual chicken detector based on machine learning (such as YOLO, R-CNN …) and then do an individual tracking of the movement of each chicken frame by frame.
​
In this project, I had a chance to build object detection models using Efficientdet, YoloV5, video annotation and conversion of yolov5 weight file to tflite format.
Kaggle Competition | Computer Vision
May 2021 ~ Sep 2021
Five times more deadly than the flu, COVID-19 causes significant morbidity and mortality. Like other pneumonias, pulmonary infection with COVID-19 results in inflammation and fluid in the lungs. COVID-19 looks very similar to other viral and bacterial pneumonias on chest radiographs, which makes it difficult to diagnose. Our computer vision model to detect and localize COVID-19 would help doctors provide a quick and confident diagnosis. As a result, patients could get the right treatment before the most severe effects of the virus take hold.
​
The goal is to implement an AI model to identify and localize COVID-19 abnormalities on chest radiographs.
For this competition, our team used EfficientV2 for classification score and Yolov5 for bounding box score. As a result, our team obtained Open Image Object Detection AP Score of 0.595 and scored public rank 493 out of 1305
Object Detection Using FasterRCNN
Internship Task | Computer Vision
3/3/2021 ~ 3/4/2021
During this 1 month internship, I had an opportunity to work as a Computer Vision & IoT intern for The Sparks Foundation.
​
The goal is to implement an object detector which identifies the classes of the objects.
​
For this project, I used TensorFlow Object Detection API and pre-trained FasterRCNN with various backbones which identifies the classes of the objects in real time webcam and Images using Google Colab and Anaconda Environment.
Pothole Detection using MaskRCNN
Personal Project | Computer Vision
3/6/2021 ~ 10/6/2021
According to Wikipedia “A pothole is a depression in a road surface, usually asphalt pavement, where traffic has removed broken pieces of the pavement”. Edmonton the “self proclaimed pothole capital” in Alberta, Canada reportedly spends $4.8 million on 450,000 potholes annually, as of 2015. In India every year approximately 1100 lives are lost to accidents caused by potholes source.
​
The goal is to implement a real-time pothole detector which can identify potholes of roads and street on images and videos.
​
For this project, I used Mask-RCNN which is an instance segmentation model, pre-trained on COCO 2017 from Tensorflow Model Zoo and using Tensorflow Object Detection API. After training 3000 steps, I obtained COCO Average Precision = 0.097, Average Recall = 0.057
Training Facebook AI's DETR on Custom Dataset
Personal Project | Computer Vision
21/6/2021 ~ 8/7/2021
DETR or DEtection TRansformer is Facebook’s newest addition to the market of available deep learning-based object detection solutions. Very simply, it utilizes the transformer architecture to generate predictions of objects and their position in an image.
​
The goal is to implement an object detector which can identify well on the faces of people especially in crowd places.
​
For this project, I've used DETR which is pretrained on COCO 2017 dataset with Resnet50 backbone and transformers acting as encoder and decoder. It took me 4:59:45 hours to finish 15 epochs with batch_size=16 using Tesla P100-PCIE. By the end of the epoch, I obtained COCO IoU Metric Box of Average Precision = 0.393 and Average Recall = 0.201
Major Advisor
School Project | Expert System
Feb 2020 - March 2020
During my 3rd year school project, I led a team of 6 members to develop an end-to-end major advisor expert system implemented in Prolog, Java and JavaFX.
​
This project was selected for 6th UIT Open Campus Project Exhibition show in 2020.
​
The goal is to advise which majors the students should take for fourth year depending on their GPA of second year, third year and according to their interest level of specific subjects.
Hotel Reservation System
School Project | Software Development
Feb 2019 ~ April 2019
During my 2nd year school project, I led a team of 6 members to develop an end-to-end hotel reservation system implemented in Java and JavaFX.
​
This project was selected for 5th UIT Open Campus Project Exhibition show in August, 2019.
​
The goal is to efficiently check-in and check-out customers, search the available rooms depending on their desire of luxury suites and view the customer list with filtration of old customers, current customers and booked customers.
​
The Food Delivery
School Project | Web Development
July 2019 ~ Sep 2019
During my 2nd year school project, I led a team of 6 members to develop a front-end website implemented in HTML, CSS and Javascript.
​
A food delivery service that searches the best restaurants for you, based on your location and customers can only order the food items from their nearby restaurants.
​
