Thisal Abeysekere

Key Skills

Software

CVAT

Label Studio

Roboflow

Top Subject Matter

No subject matter listed

Top Data Types

Medical Dicom

Geospatial Tiled Imagery

Image

Top Label Types

Segmentation

Classification

Freelancer Overview

I specialize in high-precision Computer Vision data annotation, leveraging my Master’s in Machine Learning to create researcher-grade ground truth data. My expertise lies in complex semantic segmentation and 3D volume processing, specifically within biomedical contexts (DICOM/CT scans) where I manage Hounsfield Unit calibration and RTSTRUCT data, as well as aerial imagery analysis involving high-resolution drone data. Unlike standard annotators, I utilize advanced metrics like Intersection over Union (IoU) and 3D Hausdorff Distance to audit datasets, ensuring pixel-perfect accuracy that directly optimizes model convergence." "Beyond specialized segmentation, I have extensive experience in Large-Scale Dataset Management and Quality Assurance (QA). I have curated and audited massive datasets exceeding 118,000 images, implementing rigorous data lineage pipelines and automated distribution analysis to detect class imbalances and edge cases. My technical background allows me to bridge the gap between raw data labeling and algorithmic requirements, delivering clean, structured training data ready for immediate ingestion into Supervised and Self-Supervised learning pipelines.

Labeling Experience

Lead Data Quality Specialist – Aerial Computer Vision

Label StudioGeospatial Tiled ImagerySegmentation

I led quality assurance for a high-resolution drone imagery dataset, executing ground truth verification and pixel-level semantic segmentation across 24 environmental classes. I engineered automated pixel distribution pipelines to mitigate class imbalance and ensure dataset robustness. My meticulous QA processes improved dataset quality for downstream computer vision tasks. • Audited and processed 6000x4000 aerial drone images • Applied pixel-level segmentation across 24 classes • Detected and corrected class imbalances automatically • Worked with semantic masks and RGB color masks

I led quality assurance for a high-resolution drone imagery dataset, executing ground truth verification and pixel-level semantic segmentation across 24 environmental classes. I engineered automated pixel distribution pipelines to mitigate class imbalance and ensure dataset robustness. My meticulous QA processes improved dataset quality for downstream computer vision tasks. • Audited and processed 6000x4000 aerial drone images • Applied pixel-level segmentation across 24 classes • Detected and corrected class imbalances automatically • Worked with semantic masks and RGB color masks

2024

3D Medical Imaging & Segmentation Specialist

CVATMedical DicomSegmentation

I specialized in 3D ground truth creation by managing the extraction, processing, and segmentation of 3D CT scan volumes (DICOM) with corresponding RTSTRUCT data to produce accurate 3D binary lung masks. I performed spatial normalization and rigorous QA, employing advanced visual metric auditing such as Intersection over Union (IoU), Dice Coefficient, and 3D Hausdorff Distance to guarantee accuracy. My work standardized multi-dimensional medical datasets for integration into model training pipelines. • Managed 3D DICOM and RTSTRUCT medical data for segmentation • Applied Hounsfield Unit calibration and 3D spatial resizing • Ensured pixel-perfect alignment with metric-based auditing • Optimized data for ingestion to 3D U-Net and CNN architectures

I specialized in 3D ground truth creation by managing the extraction, processing, and segmentation of 3D CT scan volumes (DICOM) with corresponding RTSTRUCT data to produce accurate 3D binary lung masks. I performed spatial normalization and rigorous QA, employing advanced visual metric auditing such as Intersection over Union (IoU), Dice Coefficient, and 3D Hausdorff Distance to guarantee accuracy. My work standardized multi-dimensional medical datasets for integration into model training pipelines. • Managed 3D DICOM and RTSTRUCT medical data for segmentation • Applied Hounsfield Unit calibration and 3D spatial resizing • Ensured pixel-perfect alignment with metric-based auditing • Optimized data for ingestion to 3D U-Net and CNN architectures

2024

Large-Scale Image Categorization Specialist

RoboflowImageClassification

I curated, organized, and verified the large ‘Food-251’ image dataset, ensuring accuracy and structural integrity for over 118,475 high-definition images. I performed extensive data cleaning, lineage tracking, and mapping of 251 specific food categories to support advanced self-supervised learning pipelines. My responsibilities included rigorous visual QA and analytical models to uncover data gaps or misclassifications. • Managed data ingestion and preparation for 251 food classes • Conducted distribution analysis to find under-represented categories • Enabled dataset usage for SimCLR self-supervision • Organized massive labeling operations for food images

I curated, organized, and verified the large ‘Food-251’ image dataset, ensuring accuracy and structural integrity for over 118,475 high-definition images. I performed extensive data cleaning, lineage tracking, and mapping of 251 specific food categories to support advanced self-supervised learning pipelines. My responsibilities included rigorous visual QA and analytical models to uncover data gaps or misclassifications. • Managed data ingestion and preparation for 251 food classes • Conducted distribution analysis to find under-represented categories • Enabled dataset usage for SimCLR self-supervision • Organized massive labeling operations for food images

2023 - 2024

Education

Work History