Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Dual-Space Augmented Intrinsic-LoRA for Wind Turbine Segmentation

Shubh Singhal, Raül Pérez-Gonzalo, Andreas Espersen, Antonio Agudo

TL;DR

This work extends Intrinsic LoRA for image segmentation, and proposes a novel dual-space augmentation strategy that integrates both image-level and latent-space augmentations, surpassing current state-of-the-art methods in WTB image segmentation.

Abstract

Accurate segmentation of wind turbine blade (WTB) images is critical for effective assessments, as it directly influences the performance of automated damage detection systems. Despite advancements in large universal vision models, these models often underperform in domain-specific tasks like WTB segmentation. To address this, we extend Intrinsic LoRA for image segmentation, and propose a novel dual-space augmentation strategy that integrates both image-level and latent-space augmentations. The image-space augmentation is achieved through linear interpolation between image pairs, while the latent-space augmentation is accomplished by introducing a noise-based latent probabilistic model. Our approach significantly boosts segmentation accuracy, surpassing current state-of-the-art methods in WTB image segmentation.

Dual-Space Augmented Intrinsic-LoRA for Wind Turbine Segmentation

TL;DR

This work extends Intrinsic LoRA for image segmentation, and proposes a novel dual-space augmentation strategy that integrates both image-level and latent-space augmentations, surpassing current state-of-the-art methods in WTB image segmentation.

Abstract

Accurate segmentation of wind turbine blade (WTB) images is critical for effective assessments, as it directly influences the performance of automated damage detection systems. Despite advancements in large universal vision models, these models often underperform in domain-specific tasks like WTB segmentation. To address this, we extend Intrinsic LoRA for image segmentation, and propose a novel dual-space augmentation strategy that integrates both image-level and latent-space augmentations. The image-space augmentation is achieved through linear interpolation between image pairs, while the latent-space augmentation is accomplished by introducing a noise-based latent probabilistic model. Our approach significantly boosts segmentation accuracy, surpassing current state-of-the-art methods in WTB image segmentation.
Paper Structure (12 sections, 5 equations, 5 figures, 2 tables)

This paper contains 12 sections, 5 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Intrinsic LoRA
  4. Segmentation-based Intrinsic LoRA (SI-LoRA)
  5. Dual-space Augmentation (DSA SI-LoRA)
  6. Experimental Results
  7. Dataset and Implementation Details
  8. Ablation Study
  9. Qualitative Evaluation
  10. Quantitative Evaluation
  11. Windfarm Dissimilarity
  12. Conclusion

Figures (5)

  • Figure 1: General schema of Segmentation-based Intrinsic LoRA (SI-LoRA) with dual-space augmentation.
  • Figure 2: Segmentation-based Intrinsic LoRA (SI-LoRA) architecture with dual-space augmentation (DSA).
  • Figure 3: Qualitative comparison of distinct data augmentation strategies. From left to right: Input image, SI-LoRA (Sec. \ref{['sec:si-lora']}), SI-LoRA with image-space augmentation (MixUp mixup), SI-LoRA with latent-space augmentation (Sec. \ref{['sec:aug']}), and SI-LoRA using both augmentations.
  • Figure 4: Boxplot test results across different windfarms.
  • Figure 5: Failure Cases. From left to right: Input Image, SI-LoRA using both augmentations (Sec. \ref{['sec:aug']}), and ground-truth segmentation masks. On both sides of the figure, the same information is displayed.