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.
Mixmatch: A holistic approach to semi-supervised learning
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.