AUTO-DIP automates the parameter selection for Deep Image Prior denoising for microscopy images. It is based on Deep Image Prior (DIP) [1] and does not require any training data. This add-on for DIP chooses parameters automatically based on microscope type, specimen, and the image itself. The code is associated with our paper [2].
-
Clone the repo and initialize submodules by running
git clone --recurse-submodules https://github.com/IPMI-ICNS-UKE/DECO-DIP
in the command line.
-
Apply patch files for the deep-image-prior repo:
(cd dip && git apply ../dip.patch) -
Set up the python virtual environment with uv:
#install uv (if not already installed) curl -LsSf https://astral.sh/uv/install.sh | sh # install packages uv sync # activate environment source .venv/bin/activate
-
Run the program. With the parameter param_path you can specify a yaml file containing the parameters. Default is ./parameters.yaml.
./main.py --param_path parameters.yaml
If you want to use MLflow for tracking, start the MLflow server before running AUTO-DIP:
mlflow server --host 127.0.0.1 --port 50000
If you want to run the program with more than one parameter file, you can specify a folder with parameter files. All yaml files in that folder are processed successively.
Example config files can be found in ./example_configs and default parameters are stored in ./default_parameters.yaml.
For detailed parameter descriptions see ./default_parameters.yaml and the parameter description in the wiki.
[1] Ulyanov D, Vedaldi A, Lempitsky V. Deep Image Prior. International Journal of Computer Vision 2020; 128 (7): 1867–88. https://doi.org/10.1007/s11263-020-01303-4.
[2] Meyer L, Wissel F, Knopp T, Pfefferle S, Fliegert R, Sandmann M, Uebler L, Möckl F, Diercks B-P, Lohr D, Werner R. Automating Parameter Selection in Deep Image Prior for Fluorescence Microscopy Image Denoising via Similarity-Based Parameter Transfer. https://doi.org/10.48550/arXiv.2601.12055.