A simplified diffusion model generates synthetic protein images, but classifiers trained on mixes of real and synthetic data fail to generalize, performing far below ResNet baselines on Kaggle test sets.
Denoising diffusion probabilistic models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Leveraging Diffusion Models for Synthetic Data Augmentation in Protein Subcellular Localization Classification
A simplified diffusion model generates synthetic protein images, but classifiers trained on mixes of real and synthetic data fail to generalize, performing far below ResNet baselines on Kaggle test sets.