France-BioImaging's Fuse My Cells challenge
Link to the challenge for more information :
- test_phase/leaderboard
- closed 07/03/2025 (initial date 28/02/2025)
- evaluation/leaderboard
- closed 17/03/2025
- Download the dataset zip files un put all of them in a folder. The following instruction need to be executed from this folder.
- Run script
01_unzip.py- Notes: the scripts for data prepration are located
in the
datafolder.
- Notes: the scripts for data prepration are located
in the
- Run script
02_tif_to_hdf5.py
Now, the zip files have been extracted to an images folder
and then put in a single file named FuseMyCells.hdf5.
usage: eval.py [-h] [--use-gpu] --method {gaussian_filter,denoise_wavelet,denoise_tv_bregman} [--args ARGS [ARGS ...]]
[--dataset DATASET] [--crop-data]
eval.py: error: the following arguments are required: --method
python eval.py --method gaussian_filter --args sigma=0.5 --dataset FuseMyCells.hdf5
Taking the docker_template given by the organizer as a start point.
from scipy import ndimage
if metadata['channel'] == 'nucleus':
image_predict = ndimage.gaussian_filter(image_input, 0.442)
else:
image_predict = ndimage.gaussian_filter(image_input, 0.5)The filter sigma values have been manually selected from evaluation on the training dataset.
The evaluation process is done using eval.py and in our case usage of the run.sh script.
- Update README (cleaning for final version)
- Add code for the docker
- Add result used to specify methods
- Update README (add method)
- Adding evaluation of method script
- usage for classical computer vision methods
- Update README (put my exp results in the idea section)
- Working on data acquisition and preprocessing
- script to unzip all the data at once
- script to convert all the images into a single HDF5 file for easier load