Research map / Adversarial robustness and security / Attacks on text and graphs
Edges to Shapes to Concepts: Adversarial Augmentation for Robust Vision
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 edge-map and shuffled-patch superposition augmentation | ResNet50 · ImageNet-1K | Accuracy@1: 76.4 -> 77.1 | Helped |
| Add edge-map and shuffled-patch superposition augmentation | ResNet101 · ImageNet-C | mCE: 69.8 -> 53.5 | Helped |
| Add edge-map and shuffled-patch superposition augmentation | ResNet152 · ImageNet-A | Accuracy@1: 5.9 -> 15.4 | Helped |
| Add edge-map and shuffled-patch superposition augmentation | ViT-S · ImageNet-1K | Accuracy@1: 74.6 -> 80.6 | Helped |
| Add edge-map and shuffled-patch superposition augmentation | ViT-B · ImageNet-A | Accuracy@1: 34.6 -> 62.9 | Helped |
| Add edge-map and shuffled-patch superposition augmentation | ViT-L · ImageNet-C | mCE: 33.1 -> 29.5 | Helped |
| Use ELEAS instead of TSD | ViT-S · ImageNet-1K | Accuracy@1: 76.4 -> 80.6 | Helped |
| Use ELEAS instead of TSD | ResNet101 · ImageNet-R | Accuracy@1: 44.3 -> 44.4 | No clear effect |
| Use ELEAS instead of TSD | ResNet152 · ImageNet-1K | Accuracy@1: 79.7 -> 79.0 | Hurt |
| Use ELEAS instead of TSD | ViT-S · ImageNet-Sketch | Accuracy@1: 32.4 -> 34.7 | Helped |
| Train with ELEAS rather than vanilla | ResNet101 · PASCAL VOC | semantic segmentation mIoU: 63.4 -> 65.9 | Helped |
| Train with ELEAS rather than vanilla | ResNet101 backbone · COCO-Val2017 | object detection mAP: 39.87 -> 41.65 | Helped |
Closest papers by meaning
- Harnessing Edge Information for Improved Robustness in Vision Transformers (AAAI 2024 · similarity 0.90)
- Shape-Texture Debiased Neural Network Training (ICLR 2021 · similarity 0.87)
- The Origins and Prevalence of Texture Bias in Convolutional Neural Networks (NeurIPS 2020 · similarity 0.87)
- Interpreting Adversarially Trained Convolutional Neural Networks (ICML 2019 · similarity 0.87)
- Understanding and Improving Robustness of Vision Transformers through Patch-based Negative Augmentation (NeurIPS 2022 · similarity 0.87)
- AugMax: Adversarial Composition of Random Augmentations for Robust Training (NeurIPS 2021 · similarity 0.87)
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.