Zhiyou Z.

Key Skills

Software

Other

OpenCV AI Kit (OAK)

Label Studio

Labelbox

Top Subject Matter

Image aesthetics assessment and explainable visual feedback

Agricultural image counting for complex scenes

Medical image segmentation and classification

Top Data Types

Image

Medical Dicom

Text

Top Task Types

Object Detection

Classification

Segmentation

Text Generation

Entity (NER) Classification

Data Collection

Freelancer Overview

I have a background in software engineering and AI vision projects, and understand the interrelationships between data collection, annotation quality, model training and performance evaluation. I have previously trained models and analysed results using the TAD66K image aesthetics dataset, and have generated image descriptions using BLIP and GPT-4o-mini to support model output review and case comparisons. Additionally, my final-year project focused on the segmentation and classification of HEp-2 IIF cell images, covering medical image sample analysis, target region identification, class classification and model evaluation. My strengths lie in my familiarity with image data, outlier analysis, evaluation metrics and quality control of AI outputs. I am proficient in using tools such as Python, SQL and Excel to support data organisation, sample screening, result review and the optimisation of annotation rules.

Labeling Experience

Apple Counting Analysis in Complex Scenes — Machine Vision Course Project (Team Lead)

OpenCV AI Kit (OAK)ImageObject Detection

Designed an agricultural computer-vision image counting solution for complex scenes involving occlusion, lighting variation, and dense fruit clusters, emphasizing how target definition and data distribution affect model performance. Built and tested a lightweight apple counting model and performed ablation studies to compare baseline and attention-augmented variants. Used Grad-CAM visualization and prediction-error distribution analysis to identify key failure sources affecting reliable counting outputs. • Defined and reviewed image counting targets under challenging conditions (occlusion, reflections, background interference) • Conducted ablation experiments (Baseline, Gate, CBAM, Full) to evaluate model components • Applied Grad-CAM and error distribution analysis to interpret mispredictions • Coordinated team task breakdown, experiment documentation, and result review as team lead

2025 - 2025

Multimodal Image Aesthetic Analysis System — AI Data and Model Evaluation (Individual Final Year Project)

OtherImage

Developed and evaluated an image aesthetic assessment pipeline using the TAD66K image aesthetics dataset to train and analyze model performance. Generated explainable image-level feedback by mapping scoring labels and text descriptions to predicted outputs, and supporting manual review with case comparison. Focused on diagnosing scoring deviations and description quality issues through sample-level analysis of model outputs. • Used image quality assessment and explainable feedback formulation based on dataset labels • Integrated BLIP and GPT-4o-mini to produce and enhance image description outputs with scores • Structured the workflow into upload, scoring, text feedback generation, and results display modules • Analyzed results with SRCC metrics and inference-time reporting plus qualitative case studies

2024 - 2025

Education

Work History