Spaces:
Runtime error
Runtime error
File size: 2,350 Bytes
96a28c6 f85e212 96a28c6 f85e212 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 |
---
title: Medfusion App
emoji: 🔬
colorFrom: pink
colorTo: gray
sdk: streamlit
sdk_version: 1.15.2
app_file: app.py
pinned: false
license: mit
---
Medfusion - Medical Denoising Diffusion Probabilistic Model
=============
Paper
=======
Please see: [**Diffusion Probabilistic Models beat GANs on Medical 2D Images**]()
*Figure: Medfusion*
  \
*Figure: Eye fundus, chest X-ray and colon histology images generated with Medfusion (Warning color quality limited by .gif)*
Demo
=============
[Link]() to streamlit app.
Install
=============
Create virtual environment and install packages: \
`python -m venv venv` \
`source venv/bin/activate`\
`pip install -e .`
Get Started
=============
1 Prepare Data
-------------
* Go to [medical_diffusion/data/datasets/dataset_simple_2d.py](medical_diffusion/data/datasets/dataset_simple_2d.py) and create a new `SimpleDataset2D` or write your own Dataset.
2 Train Autoencoder
----------------
* Go to [scripts/train_latent_embedder_2d.py](scripts/train_latent_embedder_2d.py) and import your Dataset.
* Load your dataset with eg. `SimpleDataModule`
* Customize `VAE` to your needs
* (Optional): Train a `VAEGAN` instead or load a pre-trained `VAE` and set `start_gan_train_step=-1` to start training of GAN immediately.
2.1 Evaluate Autoencoder
----------------
* Use [scripts/evaluate_latent_embedder.py](scripts/evaluate_latent_embedder.py) to evaluate the performance of the Autoencoder.
3 Train Diffusion
----------------
* Go to [scripts/train_diffusion.py](scripts/train_diffusion.py) and import/load your Dataset as before.
* Load your pre-trained VAE or VAEGAN with `latent_embedder_checkpoint=...`
* Use `cond_embedder = LabelEmbedder` for conditional training, otherwise `cond_embedder = None`
3.1 Evaluate Diffusion
----------------
* Go to [scripts/sample.py](scripts/sample.py) to sample a test image.
* Go to [scripts/helpers/sample_dataset.py](scripts/helpers/sample_dataset.py) to sample a more reprensative sample size.
* Use [scripts/evaluate_images.py](scripts/evaluate_images.py) to evaluate performance of sample (FID, Precision, Recall)
Acknowledgment
=============
* Code builds upon https://github.com/lucidrains/denoising-diffusion-pytorch
|