KrP77-HF/defect_detection

Defect Detection for Industrial Quality Control

This repository contains the datasets and model checkpoints for an automated quality control system designed to detect defects in plastic injection-molded components.

🔍 Project Overview

This project applies deep learning to industrial manufacturing. By utilizing YOLOv8 architectures, the system identifies surface defects in real-time, specifically optimized for deployment on edge devices like the Raspberry Pi 5.

📂 Repository Structure

• augmented_data: Datasets expanded with geometric and color augmentations.
• augmented_data_v2: Datasets expanded with geometric and color augmentations and then cropped with yolo model.
• cvat_exports / label_studio_exports / label_studio_exports_2000: Annotated datasets in YOLO format.
• dataset: type_A and type_B with raw backup and splits.
• models / runs: Exported weights (.pt) and training logs.
• pi_camera_dataset: Real-world test images captured in a factory environment.
• unannotated_segmentation: Training/testing data for unannotated segmentation.
• yolo_training: Yolo training data with labels and images.

💻 GitHub Source Code

The full implementation, including training pipelines, data preprocessing scripts, and deployment code, is available on GitHub:

👉 GitHub: Defect-Detection-Project

---

Quick Start

To download the data using the Hugging Face CLI:

huggingface-cli download KrP77-HF/defect_detection --repo-type dataset --local-dir ./data