Research map / Image generation / Autoregressive and latent generation
Expanding Small-Scale Datasets with Guided Imagination
Yifan Zhang, Daquan Zhou, Bryan Hooi, Kai Wang, Jiashi Feng
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 |
|---|---|---|---|
| expand dataset 20x with GIF-SD instead of training on the original small data | ResNet-50 · six natural image datasets | average accuracy: 30.9 -> 67.8 | Improved |
| expand dataset 5x with GIF-SD | ResNet-50 · three medical datasets | average accuracy: 68.2 -> 81.7 | Improved |
| use GIF-SD instead of RandAugment expansion | ResNet-50 · natural image datasets | average accuracy: 51.4 -> 67.8 | Improved |
| add guidance (GIF-SD) over unguided Stable Diffusion expansion | GIF-SD · natural image datasets | average accuracy: 54.3 -> 67.8 | Improved |
| add guidance (GIF-DALLE) over unguided DALL-E2 expansion | GIF-DALLE · natural image datasets | average accuracy: 57.0 -> 60.8 | Improved |
| add guidance (GIF-MAE) over unguided MAE expansion | GIF-MAE · natural image datasets | average accuracy: 44.1 -> 54.4 | Improved |
| fine-tune SD on medical data before expansion | GIF-SD · medical datasets | average accuracy: 73.2 -> 81.7 | Improved |
| fine-tune DALL-E2 on medical data before expansion | GIF-DALLE · medical datasets | average accuracy: 71.4 -> 80.5 | Improved |
| use zero-shot CLIP guidance instead of random-initialized ResNet50 guidance | GIF-MAE · OrganSMNIST | accuracy: 79.0 -> 80.6 | Improved |
| use fine-tuned CLIP guidance instead of zero-shot CLIP (ablation) | GIF-MAE · OrganSMNIST | accuracy: 80.6 -> 80.7 | No clear effect |
| train on GIF-SD expanded data vs fine-tuning CLIP | ResNet-50 · PathMNIST | accuracy: 78.4 -> 86.9 | Improved |
| fine-tune CLIP on 5x GIF-SD expanded dataset | CLIP ViT-B/32 · CIFAR100-S | ID accuracy: 75.2 -> 79.4 | Improved |
Closest papers by meaning
- Distribution-Aware Data Expansion with Diffusion Models (NeurIPS 2024 · similarity 0.84)
- FlashEval: Towards Fast and Accurate Evaluation of Text-to-image Diffusion Generative Models (CVPR 2024 · similarity 0.84)
- Increasing the Utility of Synthetic Images through Chamfer Guidance (NeurIPS 2025 · similarity 0.84)
- ImagineFSL: Self-Supervised Pretraining Matters on Imagined Base Set for VLM-based Few-shot Learning (CVPR 2025 · similarity 0.83)
- MGD$^3$ : Mode-Guided Dataset Distillation using Diffusion Models (ICML 2025 · similarity 0.83)
- Low-Biased General Annotated Dataset Generation (CVPR 2025 · similarity 0.83)
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.
