Research map / Image generation / Autoregressive and latent generation
OntoAug: Rethinking Generative Data Augmentation via Ontology Guidance
Shuo Wang, Zhichuan Wang, Jun Luo
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 OntoAug generative augmentation versus vanilla training | ResNet-50 (from scratch) · CUB | top-1 accuracy (%): 65.50 -> 84.62 | Improved |
| add layout structure augmentation on top of background augmentation (ablation) | OntoAug · CUB (ResNet-50) | top-1 accuracy (%): 81.53 -> 84.62 | Improved |
| add translation (L3) layout alone (ablation) | OntoAug background-only · CUB (ResNet-50) | top-1 accuracy (%): 81.53 -> 82.29 | Improved |
| use generated backgrounds instead of Places backgrounds (ablation) | OntoAug · CUB (ResNet-50) | top-1 accuracy (%): 77.17 -> 84.62 | Improved |
| use diverse background vocabulary instead of habitat-only terms (ablation) | OntoAug · Waterbirds | average accuracy (%): 73.48 -> 74.82 | Improved |
| combine OntoAug with SnapMix instead of Diff-Mix with SnapMix | ResNet-50 (ImageNet-1K pretrained) · CUB | top-1 accuracy (%): 87.28 -> 88.97 | Improved |
| add Diff-Mix to SnapMix | ResNet-50 (ImageNet-1K pretrained) · CUB | top-1 accuracy (%): 87.75 -> 87.28 | Hurt |
| add OntoAug on ViT-B/16 | ViT-B/16 · CUB | accuracy (%): 89.37 -> 90.78 | Improved |
| apply CutMix augmentation in 10-shot CUB | ResNet-18/ResNet-50/DenseNet-121 · CUB 10-shot | average test accuracy (%): 31.79 -> 23.36 | Hurt |
| add OntoAug augmentation in 10-shot learning | three backbones · CUB 10-shot | average test accuracy (%): 31.79 -> 67.55 | Improved |
| add OntoAug data to GRPO fine-tuning | Qwen2.5-VL-3B · CUB 8-shot | accuracy: 59.51 -> 66.24 | Improved |
| add OntoAug for CAM-based localization | CAM · CUB WSOL | IoU50 localization accuracy (%): 50.85 -> 60.54 | Improved |
Closest papers by meaning
- Seeing the Unseen: Visual Common Sense for Semantic Placement (CVPR 2024 · similarity 0.85)
- MMMG: A Massive, Multidisciplinary, Multi-Tier Generation Benchmark for Text-to-Image Reasoning (NeurIPS 2025 · similarity 0.85)
- Salient Concept-Aware Generative Data Augmentation (NeurIPS 2025 · similarity 0.85)
- Localizing and Editing Knowledge In Text-to-Image Generative Models (ICLR 2024 · similarity 0.84)
- Semantic-Guided Generative Image Augmentation Method with Diffusion Models for Image Classification (AAAI 2024 · similarity 0.84)
- Efficient bias mitigation in T2I diffusion models using Concept Graphs (NeurIPS 2026 · similarity 0.84)
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.
