Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Sim-to-Real Domain Adaptation for Deformation Classification

Joel Sol, Jamil Fayyad, Shadi Alijani, Homayoun Najjaran

TL;DR

The paper tackles deformation classification by bridging the sim-to-real gap through CASNet, a content-aware, disentangled network that transfers synthetic deformation images into realistic-like appearances using CADT. By generating a large synthetic dataset in Blender and employing CASNet to align content and style across domains, the approach yields substantially better sim-to-real classification than CycleGAN baselines, while requiring minimal unlabeled real data. The workflow combines a DRANet-inspired architecture, perceptual guidance from VGG-19, and content-adaptive transfer to suppress artifacts, achieving improved deformation detection performance with practical implications for structural health monitoring and automated quality control. Future directions include depth-enabled data, transformer-based architectures, and refined style losses such as SWD to further close the sim-to-real gap.

Abstract

Deformation detection is vital for enabling accurate assessment and prediction of structural changes in materials, ensuring timely and effective interventions to maintain safety and integrity. Automating deformation detection through computer vision is crucial for efficient monitoring, but it faces significant challenges in creating a comprehensive dataset of both deformed and non-deformed objects, which can be difficult to obtain in many scenarios. In this paper, we introduce a novel framework for generating controlled synthetic data that simulates deformed objects. This approach allows for the realistic modeling of object deformations under various conditions. Our framework integrates an intelligent adapter network that facilitates sim-to-real domain adaptation, enhancing classification results without requiring real data from deformed objects. We conduct experiments on domain adaptation and classification tasks and demonstrate that our framework improves sim-to-real classification results compared to simulation baseline.

Sim-to-Real Domain Adaptation for Deformation Classification

TL;DR

The paper tackles deformation classification by bridging the sim-to-real gap through CASNet, a content-aware, disentangled network that transfers synthetic deformation images into realistic-like appearances using CADT. By generating a large synthetic dataset in Blender and employing CASNet to align content and style across domains, the approach yields substantially better sim-to-real classification than CycleGAN baselines, while requiring minimal unlabeled real data. The workflow combines a DRANet-inspired architecture, perceptual guidance from VGG-19, and content-adaptive transfer to suppress artifacts, achieving improved deformation detection performance with practical implications for structural health monitoring and automated quality control. Future directions include depth-enabled data, transformer-based architectures, and refined style losses such as SWD to further close the sim-to-real gap.

Abstract

Deformation detection is vital for enabling accurate assessment and prediction of structural changes in materials, ensuring timely and effective interventions to maintain safety and integrity. Automating deformation detection through computer vision is crucial for efficient monitoring, but it faces significant challenges in creating a comprehensive dataset of both deformed and non-deformed objects, which can be difficult to obtain in many scenarios. In this paper, we introduce a novel framework for generating controlled synthetic data that simulates deformed objects. This approach allows for the realistic modeling of object deformations under various conditions. Our framework integrates an intelligent adapter network that facilitates sim-to-real domain adaptation, enhancing classification results without requiring real data from deformed objects. We conduct experiments on domain adaptation and classification tasks and demonstrate that our framework improves sim-to-real classification results compared to simulation baseline.
Paper Structure (13 sections, 3 equations, 5 figures, 2 tables)

This paper contains 13 sections, 3 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. Data Generation
  5. CycleGAN
  6. Content-Aware Sim-to-real Network (CASNet)
  7. VGG-16 Classifier
  8. Training
  9. CycleGAN
  10. CASNet
  11. VGG-16 Classifier
  12. Results and Discussion
  13. Conclusion

Figures (5)

  • Figure 1: CASNet Architecture
  • Figure 2: CycleGAN Image Analysis
  • Figure 3: Domain Adaptation Results from CASNet
  • Figure 4: CASNet Generated Images PCA Visualizations
  • Figure 5: Confusion matrices for Comparing CASNet Performance