Reproduce “AugDiff: Diffusion based Feature Augmentation for Multiple Instance Learning in Whole Slide Image” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhuchen Shao, Liuxi Dai, Yifeng Wang, Haoqian Wang, Yongbing Zhang
AugDiff presents a Diffusion Model-based framework for augmenting features in Multiple Instance Learning (MIL) for Whole Slide Image (WSI) analysis, achieving superior model performance and processing speeds over 30 times faster than traditional image augmentation by generating diverse, semantically consistent features.
Key insights
- Diffusion Models can effectively generate diverse and semantically consistent feature augmentations, overcoming the limitations of prior generative methods like GANs and Mixup for WSI analysis.
- Guiding Diffusion Model training with features derived from traditional image augmentations enables learning high-quality augmentation directions that improve MIL model generalization.
- A carefully designed two-stage K-step diffusion and K-step denoising strategy in the feature space is crucial for balancing the introduction of diversity with the preservation of original semantic information.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.