Research map / Label-efficient and robust learning / Semi-supervised learning
Adversarial Data Augmentation via Deformation Statistics
Sahin Olut, Zhengyang Shen, Zhenlin Xu, Samuel Gerber, Marc Niethammer
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add adversarial direction to EigAug (rand r) (ablation) | EigAug · OAI knee MRI, 4 training samples | Dice (%): 73.3 -> 78.4 | Improved |
| add adversarial direction to EigAug (rand r) (ablation) | EigAug · OAI knee MRI, 8 training samples | Dice (%): 76.2 -> 81.0 | Improved |
| add adversarial direction to EigAug (fixed r=2) (ablation) | EigAug · OAI knee MRI, 4 training samples | Dice (%): 73.6 -> 76.1 | Improved |
| add adversarial direction to EigAug (fixed r=2) (ablation) | EigAug · OAI knee MRI, 8 training samples | Dice (%): 77.7 -> 78.7 | Improved |
| randomize deformation magnitude r~U(1.5,3) instead of fixed r=2 (ablation) | AdvEigAug · OAI knee MRI, 4 training samples | Dice (%): 76.1 -> 78.4 | Improved |
| randomize deformation magnitude r~U(1.5,3) instead of fixed r=2 (ablation) | AdvEigAug · OAI knee MRI, 8 training samples | Dice (%): 78.7 -> 81.0 | Improved |
| use fixed r=3 instead of r=2 (ablation) | AdvEigAug · OAI knee MRI, 8 training samples | Dice (%): 78.7 -> 76.9 | Hurt |
| use LDDMM deformation model instead of displacement field network (ablation) | AdvEigAug rand r · OAI knee MRI, 4 training samples | Dice (%): 78.4 -> 78.5 | Improved |
| use online instead of offline augmentation with LDDMM (ablation) | L-AdvEigAug rand r · OAI knee MRI, 4 training samples | Dice (%): 78.5 -> 79.1 | Improved |
| use online instead of offline augmentation with displacement model (ablation) | AdvEigAug rand r · OAI knee MRI, 8 training samples | Dice (%): 81.0 -> 81.1 | No clear effect |
| use adversarial statistical-deformation augmentation instead of AdvAffine | AdvAffine · OAI knee MRI, 4 training samples | Dice (%): 75.1 -> 79.1 | Improved |
| add adversarial AdvEigAug augmentation vs no augmentation | 3D U-Net · OAI knee MRI, 4 training samples | Dice (%): 75.2 -> 79.1 | Improved |
Closest papers by meaning
- Data Augmentation Using Learned Transformations for One-Shot Medical Image Segmentation (CVPR 2019 · similarity 0.88)
- Modeling the Probabilistic Distribution of Unlabeled Data for One-shot Medical Image Segmentation (AAAI 2021 · similarity 0.87)
- AIM: An Auto-Augmenter for Images and Meshes (CVPR 2022 · similarity 0.87)
- Deep Complementary Joint Model for Complex Scene Registration and Few-shot Segmentation on Medical Images (ECCV 2020 · similarity 0.87)
- Augmentation-invariant Learning Strategy via Data Augmentation for Improving Model Generalization (AAAI 2026 · similarity 0.86)
- Metric Learning for Image Registration (CVPR 2019 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
