Research map / Label-efficient and robust learning / Recognition and long-tail
VITA: A Multi-Source Vicinal Transfer Augmentation Method for Out-of-Distribution Generalization
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 |
|---|---|---|---|
| Use VITA instead of AugMix | ImageNet training with ResNet-50 · ImageNet-C | mCE: 68.4 -> 52.1 | Helped |
| Use VITA instead of AugMix | CIFAR-10 training across four architectures · CIFAR-10-C | mCE: 12.5 -> 9.7 | Helped |
| Use VITA instead of AugMix | CIFAR-100 training across four architectures · CIFAR-100-C | mCE: 38.3 -> 34.4 | Helped |
| Add VITA-generated samples to adversarial training | WRN-34-10 with TRADES-based training, β = 1.0 · CIFAR-10 | FGSM accuracy: 61.00 -> 68.02 | Helped |
| Add VITA-generated samples to adversarial training | WRN-34-10 with TRADES-based training, β = 6.0 · CIFAR-10 | AutoAttack accuracy: 53.27 -> 54.35 | Helped |
| Transfer vicinal differences from weak augmentations (ablation) | AllConvNet · CIFAR-10-C | error rate | Helped |
| Train and infer the translator without transferred vicinal differences (ablation) | VITA with AllConvNet · CIFAR-10-C | error rate | Hurt |
| Use Gaussian noise instead of transferred vicinal differences as translator input (ablation) | VITA with AllConvNet · CIFAR-10-C | error rate | Hurt |
| Remove VITA-generated samples from multi-source robust training (ablation) | AllConvNet · CIFAR-10-C | error rate | Hurt |
| Train with adversarially perturbed and VITA-generated samples, excluding weakly augmented samples (ablation) | AllConvNet · CIFAR-10-C | error rate | Helped |
| Train with weakly augmented and VITA-generated samples, excluding shuffled adversarial perturbations (ablation) | AllConvNet · CIFAR-10-C | error rate | Mixed |
Closest papers by meaning
- Adaptive Sensitivity Analysis for Robust Augmentation against Natural Corruptions in Image Segmentation (ICML 2025 · similarity 0.83)
- Manifold Induced Biases for Zero-shot and Few-shot Detection of Generated Images (ICLR 2025 · similarity 0.83)
- Consistency Purification: Effective and Efficient Diffusion Purification towards Certified Robustness (NeurIPS 2024 · similarity 0.82)
- Detecting Generated Images by Fitting Natural Image Distributions (NeurIPS 2025 · similarity 0.82)
- Mint: A Simple Test-Time Adaptation of Vision-Language Models against Common Corruptions (NeurIPS 2025 · similarity 0.82)
- HYBOOD: A Hybrid Generative Model for Out-of-Distribution Detection with Corruption Estimation (AAAI 2025 · similarity 0.82)
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.