Research map / Image generation / Fast diffusion sampling
Active Generation for Image Classification
Tao Huang, Jiaqi Liu, Shan You, Chang Xu
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 |
|---|---|---|---|
| generate images guided by misclassified validation samples (ActGen) instead of random fixed-prompt generation with the same budget (ablation) | ResNet-50 on ImageNet · ImageNet-1k val | top-1 accuracy (%): 76.64 -> 78.65 | Improved |
| add image guidance from real hard images (ablation) | random SD generation · ImageNet-1k val (ResNet-50) | top-1 accuracy (%): 76.64 -> 77.93 | Improved |
| add attention-mask selective guidance (AIG) (ablation) | ActGen image guidance · ImageNet-1k val (ResNet-50) | top-1 accuracy (%): 77.93 -> 78.15 | Improved |
| add contrastive loss against memory bank (ablation) | ActGen · ImageNet-1k val (ResNet-50) | top-1 accuracy (%): 78.15 -> 78.36 | Improved |
| add adversarial classification loss (ablation) | ActGen · ImageNet-1k val (ResNet-50) | top-1 accuracy (%): 78.36 -> 78.65 | Improved |
| increase validation set size from 0.1K/1K to 5K or more (ablation) | ActGen · ImageNet-1k val (ResNet-50) | top-1 accuracy (%) | Improved |
| use ActGen with 10% generated images instead of 94% generated images (Azizi et al.) | ResNet-50 on ImageNet · ImageNet-1k val | top-1 accuracy (%): 78.17 -> 78.65 | Improved |
| apply ActGen on CIFAR-100 with ResNet-50 | ActGen · CIFAR-100 (ResNet-50) | top-1 accuracy (%): 77.06 -> 77.33 | No clear effect |
| use ActGen instead of Real filtering at 1-shot on EuroSAT | few-shot EuroSAT · EuroSAT 1-shot (2K generated) | accuracy (%): 70.32 -> 66.78 | Hurt |
| increase synthetic images beyond about 1,850 for few-shot Pets baseline (ablation) | Real guidance baseline · Pets 1-shot | accuracy (%): 87.65 -> 87.41 | Hurt |
| reimplement Azizi et al. on Stable Diffusion rather than Imagen | Azizi et al. · ImageNet-1k val (ResNet-50) | top-1 accuracy (%): 78.17 -> 76.83 | Hurt |
Closest papers by meaning
- Leverage Class-Specific Accuracy to Guide Data Generation for Improving Image Classification (ICML 2024 · similarity 0.85)
- Beyond Generation: A Diffusion-based Low-level Feature Extractor for Detecting AI-generated Images (CVPR 2025 · similarity 0.84)
- A Generative Approach at the Instance-Level for Image Segmentation Under Limited Training Data Conditions (Student Abstract) (AAAI 2025 · similarity 0.84)
- ADDP: Learning General Representations for Image Recognition and Generation with Alternating Denoising Diffusion Process (ICLR 2024 · similarity 0.84)
- DiverGen: Improving Instance Segmentation by Learning Wider Data Distribution with More Diverse Generative Data (CVPR 2024 · similarity 0.84)
- Generative Active Learning for Long-tailed Instance Segmentation (ICML 2024 · 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.
