This is the code for this blogpost detailing my journey attempting to reproduce the U-Net paper from 2015.
uv venv
source .venv/bin/activate
python -m ensurepip
pip3 install -r requirements.txt
pip3 install -r requirements.dev.txt
pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu128Check out / change hyperparams in src/train.py, as well as the wandb project / organization.
uv run python src/train.pyAll datasets are vendored here, but due credit and lineage is below.
To preview a dataloader (with transforms and so on):
PYTHONPATH=src uv run python src/dataset/__init__.py phc # phc, em, or helaThe dataset contains 30 ssTEM (serial section Transmission Electron Microscopy) images taken from the Drosophila larva ventral nerve cord (VNC). The images represent a set of consecutive slices within one 3D volume. Corresponding segmentation ground truths are also provided in this dataset.
Original URL is down (archived version is available here), but Hoang Pham uploaded it to Github.
As far as I can dig, the dataset had originally been published in this paper.
PHC-C2DH-U373 Trainset URL
Glioblastoma-astrocytoma U373 cells on a polyacrylamide substrate
Dr. S. Kumar. Department of Bioengineering, University of California at Berkeley, Berkeley CA (USA)
DIC-C2DH-HeLa Trainset URL
HeLa cells on a flat glass
Dr. G. van Cappellen. Erasmus Medical Center, Rotterdam, The Netherlands