This project is a tool for detecting cells in medical images. It is based on the following Kaggle challenge.
## Project structure
TODO: Add structure
For this, run the following command:
conda install --file environement.txt
Then, run:
python setup.py develop
## Workflow
- Get and process the data
- Train the model
- Postprocess the predictions
Once you have installed the project dependencies, you will have access to the Kaggle official CLI. Run kaggle --help to confirm this.
To get the data, run:
kaggle competitions download -c data-science-bowl-2018
You will need to accept the competition conditions and create an API key first.
## Running TensorBoard
To run TensorBoard (a great visualization tool),
tensorboard --logdir=/path/to/tb_logs
## Tips
- It is better to use skimage instead of numpy for reading and processing images.
- It is even better to use Keras built-in image processing capabilities.
- Log the various image sizes when debugging your data processing pipeline.
- Use only one image when debugging your data pipeline (so that to avoid loading all the data multiple times).
## Sources and useful links
- U-net Kaggle kernel: https://www.kaggle.com/keegil/keras-u-net-starter-lb-0-277
- U-net paper: https://arxiv.org/abs/1505.04597
- Upsampling basics: https://www.cs.toronto.edu/~guerzhoy/320/lec/upsampling.pdf