A dual text encoder framework for augmenting pothole image-mask pairs using diffusion models

Le résumé fourni par la source

This paper proposes a data augmentation method based on a dual-text encoding framework to enhance pothole detection datasets. The method introduces dual prompts to generate both mask images and pothole images, using the generated mask images as control conditions to guide the generation of pothole images. Finally, data augmentation experiments were conducted on the Pothole Segmentation dataset, and the augmented dataset was used to train segmentation models such as FPN, PSPNet, and U-Net. Experimental results indicate that this method significantly improves pothole detection performance, with the IoU metric increasing by 0.5 to 1.5 percentage points. These findings demonstrate the potential of leveraging dual-text prompts for effective data augmentation in improving model accuracy in real-world applications.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.