Model weights and data for paper "Self-Supervised Learning Approach for Multi-label Sewer Defect Classification"

<strong>Repository for model weights and data</strong> accompanying the paper <strong>“Self-Supervised Learning Approach for Multi-label Sewer Defect Classification”</strong> by <strong>Tugba Yildizli, Tianlong Jia, Jeroen Langeveld, and Riccardo Taormina</strong>.<br>This repository provides:<strong>SwAV pre-trained weights</strong> for self-supervision,<strong>Fine-tuned model weights</strong> (fully supervised and semi-supervised),<strong>Supporting data/configs</strong> used to train and analyze these models.<br>Researchers can (i) <strong>fine-tune</strong> the SwAV pre-trained backbones on their own sewer datasets for semi-supervised learning, and (ii) <strong>evaluate</strong> our fine-tuned models for reproducibility. All code examples use <strong>PyTorch</strong>.本仓库为Tugba Yildizli、Tianlong Jia、Jeroen Langeveld与Riccardo Taormina合著的论文《面向多标签管道缺陷分类的自监督学习方法》（"Self-Supervised Learning Approach for Multi-label Sewer Defect Classification"）的配套模型权重与数据集仓库。 本仓库提供以下内容：用于自监督学习的SwAV预训练权重、涵盖全监督与半监督两种训练范式的微调模型权重，以及用于训练与分析上述模型的配套数据与配置文件。 研究人员可借助本仓库完成两项工作：(i) 在自有管道缺陷数据集上对SwAV预训练骨干网络进行微调以开展半监督学习研究；(ii) 对我们的微调模型进行评估以实现实验复现。所有代码示例均基于PyTorch框架实现。