You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This project aims to use Generative models to rebalance imbalanced datasets. We compare Stable Diffusion to GANs and other more traditional data augmentation methods.
Images from GAN
DCGAN with slightly modified architecture
Stable Diffusion + Active Learning
Off-the-shelf Stable Diffusion w. Monte Carlo Dropout
Results
Empirical results from classification after training on ResNet18 w. different datasets
PCA on features in Resnet18 with the full, Stable Diffusion-created and Stable Diffusion + active learning created datasets