Diffusion-based foundation model for 5 task(DF5T)

User Interface for DF5T

1. Download the Finetuned Models

You can download the finetuned models and the examples of DF5T from https://huggingface.co/dumengze/DF5T.

tasks/
    deno_em/
    sr2/
    deblur_em/
    inpaint_em/
    isotropic_em/

2. Install Packages

• Python 3.8
• Pytorch 2.0.1, CUDA 11.7 and CUDNN
• Python Packages:

You can install the required python packages by the following command:

conda create -n DF5T python=3.8
conda activate DF5T
pip install -r requirements.txt

3. Run the Web Interface

You can run the interactive software with the following command:

python app.py

You can learn how to use the app through https://youtu.be/GrUuY4oedGE. See the instructions in the image below.

Train

python /guided-diffusion-main/scripts/image_train.py

Test

1. Prepare the datasets

1. All training and test data involved in the experiments are publicly available datasets. You can download our preprocessed data from Science Data Bank, ScienceDB and unzip them into the corresponding folders.

2. If your data is three-dimensional, preprocess it using the following code:

python pre/train.py

3. To preprocess 2D data, use the following code:

python pre/enhance.py

4. The dataset structure should be as follows:

exp
    datasets
            MitEM/MitEM

5. Use the following code to implement each task:

Denoise

python main.py --ni --config DF5T_512.yml --timesteps 50 --deg deno_em --sigma_0 0.2

Super resolution

python main.py --ni --config DF5T_512.yml --timesteps 50 --deg sr2 --sigma_0 0.2

Deblur

python main.py --ni --config DF5T_512.yml --timesteps 50 --deg deblur_em --sigma_0 0.2

2D-Inpaint

python main.py --ni --config DF5T_512.yml --timesteps 50 --deg inp_em --sigma_0 0.2

3D-Isotropic

python main.py --ni --config DF5T_512.yml --timesteps 50 --deg isotropic_em --sigma_0 0.2