Insulator Dataset Introduction

This dataset utilizes domain randomization techniques involving equipment structure, texture information, scene details, and camera rendering poses. An automated method is established for generating insulator defect data, including the creation of defect labeling criteria and automated annotation generation. Within a three-dimensional space, the dataset constructs parametrically adjustable insulator disc models using procedural modeling, alongside parameterized material texture models. Defect modules are generated through noise-based textures to simulate insulator disc damages. The dataset's diversity is enhanced by introducing background objects and occlusion structures in the environment. Domain randomization is employed to adjust model structures and scene details. Methods for camera poses and rendering parameter settings are established, along with an automated annotation evaluation process for defect data. Ultimately, this enables the batch generation of insulator defect data.

Based on dozens of transmission towers, this dataset generates tens of thousands of insulator damage models by randomly adding multiple points, forms, backgrounds (with hundreds of background photos), and numbers/types of damaged insulators. The randomly generated insulator string damage models cover two material types: glass and ceramic, as well as various numbers and types of damaged insulator discs, including broken defects, drop (explosion). Academic researchers can freely download 4000 images of damaged insulator strings after registering and logging in. Commercial users can download the complete dataset (including 100,000 images of insulator damage) after registering and receiving approval upon submission of their enterprise registration certificate.