AI Paper Atlas

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

NeurIPS 2023 · Image generation › Autoregressive and latent generation · approach: Synthetic training data from generators

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 changedWhereMeasuredEffect
expand dataset 20x with GIF-SD instead of training on the original small dataResNet-50 · six natural image datasetsaverage accuracy: 30.9 -> 67.8Improved
expand dataset 5x with GIF-SDResNet-50 · three medical datasetsaverage accuracy: 68.2 -> 81.7Improved
use GIF-SD instead of RandAugment expansionResNet-50 · natural image datasetsaverage accuracy: 51.4 -> 67.8Improved
add guidance (GIF-SD) over unguided Stable Diffusion expansionGIF-SD · natural image datasetsaverage accuracy: 54.3 -> 67.8Improved
add guidance (GIF-DALLE) over unguided DALL-E2 expansionGIF-DALLE · natural image datasetsaverage accuracy: 57.0 -> 60.8Improved
add guidance (GIF-MAE) over unguided MAE expansionGIF-MAE · natural image datasetsaverage accuracy: 44.1 -> 54.4Improved
fine-tune SD on medical data before expansionGIF-SD · medical datasetsaverage accuracy: 73.2 -> 81.7Improved
fine-tune DALL-E2 on medical data before expansionGIF-DALLE · medical datasetsaverage accuracy: 71.4 -> 80.5Improved
use zero-shot CLIP guidance instead of random-initialized ResNet50 guidanceGIF-MAE · OrganSMNISTaccuracy: 79.0 -> 80.6Improved
use fine-tuned CLIP guidance instead of zero-shot CLIP (ablation)GIF-MAE · OrganSMNISTaccuracy: 80.6 -> 80.7No clear effect
train on GIF-SD expanded data vs fine-tuning CLIPResNet-50 · PathMNISTaccuracy: 78.4 -> 86.9Improved
fine-tune CLIP on 5x GIF-SD expanded datasetCLIP ViT-B/32 · CIFAR100-SID accuracy: 75.2 -> 79.4Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy