Multi-label sewer defect classification based on CLIP with fine-to-coarse contextual representations

Key Points

• The approach achieved higher accuracy in multi-label classification of sewer defects, improving detection reliability.
• Accuracy increased by 15% for complex defects when contextual representations were utilized within the CLIP model.
• Deep learning methods were employed to process and classify the sewer images effectively, demonstrating robust performance.
• Suggests a need for further enhancements in contextual representation to uncover even deeper insights from sewer inspection data.