Generating Progressive Images from Pathological Transitions via Diffusion Model
Zeyu Liu, Tianyi Zhang, Yufang He, Yu Zhao, Yunlu Feng, Guanglei Zhang
Arxiv, GitHub
Pathological image analysis is a crucial field in deep learning applications. However, training effective models demands large-scale annotated data, which faces challenges due to sampling and annotation scarcity. The rapid developing generative models show potential to generate more training sam-ples in recent studies. However, they also struggle with generalization diver-sity when limited training data is available, making them incapable of gener-ating effective samples. Inspired by pathological transitions between differ-ent stages, we propose an adaptive depth-controlled diffusion (ADD) net-work for effective data augmentation. This novel approach is rooted in do-main migration, where a hybrid attention strategy blends local and global at-tention priorities. With feature measuring, the adaptive depth-controlled strategy guides the bidirectional diffusion. It simulates pathological feature transition and maintains locational similarity. Based on a tiny training set (samples ≤ 500), ADD yields cross-domain progressive images with corre-sponding soft labels. Experiments on two datasets suggest significant im-provements in generation diversity, and the effectiveness of the generated progressive samples is highlighted in downstream classification tasks.
- Train a diffusion model on your data based on the guided-diffusion
- Assign the path of trained models, and then generate intermediate images
(The total diffusion process includes 1000 steps, and we can get 10 intermediate images)
python scripts/frequency_generating_m_samples.py --diffusion_steps=1000 --amount=10This implementation is based on / inspired by:
openai/guided-diffusion
openai/improved-diffusion
suxuann/ddib
Same as IDDPM / ADM / DDIB
The comparison methods are listed here:
| Model | Based method | Paper | Code |
|---|---|---|---|
| ProGAN | GAN | Progressive Growing of GANs for Improved Quality, Stability, and Variation | Github |
| IDDPM | Diffusion | Improved Denoising Diffusion Probabilistic Models | Github |
| LoFGAN | GAN | LoFGAN: Fusing Local Representations for Few-shot Image Generation | Github |
| MixDL | GAN | Towards Expert-Level Medical Question Answering with Large Language Models | Github |