As an early stage researcher, I am engaged in interdisciplinary research within the GAP Project, focusing on topological analysis of bone medical images. Here are some key highlights:
- Standard Bone Morphometry
- Extracted bone morphometric measurements using BoneJ (a plugin for ImageJ).
- Topological Data Analysis
- Generated vector representations of the cubical persistent homology computed from the bone medical images.
- Traditional Machine Learning
- Trained machine learning models for regression to predict bone strength using extracted features.
Embarking my journey as a CIP Fellow stationed at the QuICMaPP Facility, I have actively participated in a diverse range of tasks and initiatives, leveraging my skills and expertise in the field. Here are some key highlights:
- Object Detection Model Training and Hyperparameter Tuning
- Trained YOLOv5 and YOLOv8 models using a comprehensive dataset consisting of field and laboratory-generated images featuring plastic waste.
- Conducted hyperparameter tuning on the trained YOLOv5 model.
- Performed five-fold cross validation to estimate the performance of the YOLO models on new data.
- Class Activation Mapping (CAM) Analysis with the Trained YOLO Models
- Applied Eigen-CAM to identify crucial regions of interest within the output of the YOLOv5 models and prepared a similar implementation on YOLOv8.
- Observed the CAM images to prepare a detailed analysis on the interpretations of the models.
- Performance Evaluation of the Models
- Summarized the performance metrics and created visualizations to comopare the YOLOv5 (Default and Optimized) models and the YOLOv8 model.
- Provided a short discussion of the history of YOLO, hyperparameter evolution, and object detection metrics.
- Used hypothesis testing to determine whether the models are significantly difference from each other.
- Literature Review on Image Monitoring Tools
- Contributed to a comprehensive literature review on image monitoring tools focused on macroplastic pollution.
- Assisted in identifying and evaluating open-source datasets relevant to the field.
- Co-Supervision on Undergraduate Thesis
- Provided comments on the methodology and the results to improve the quality of the study.
- Suggested improvements on the writing of a thesis project centered around model-assisted labeling.
- Read advanced concepts on robust estimation and hypothesis testing.
- Co-Supervision on the Student Assistantship (SA) Program
- Conducted initial assessment of the applicants to the SA program.
- Conducted interviews to the shortlisted applicants.
- Supervised accepted SAs in macroplastic tasks within the facility.
- Christmas Lantern Project
- Successfully deployed the object detection model onto a compact Raspberry Pi device.
- Integrated a light show into a plastic pollution-themed Christmas lantern, showcasing the versatility of the model in real-world applications.
- Citizen Science Event
- Investigated the feasibility of utilizing an annotated dataset from the public to develop a reliable plastic object detection model.
- Designed posters and signages for the conduct of the event catered to the UP Baguio community and an exclusive event for high school students outside UP Baguio.
These accomplishments reflect my commitment to advancing knowledge and applying cutting-edge technologies in meaningful and creative ways. I am excited to continue contributing to innovative projects and driving positive change in my field.
- Ignacio, P.S., Bulauan, J.-A., & Manzanares, J.R. (2020). A Topology Informed Random Forest Classifier for ECG Classification. In 2020 Computing in Cardiology Conference (CinC). 2020 Computing in Cardiology Conference. Computing in Cardiology. https://doi.org/10.22489/cinc.2020.297