burke_leaf_veins

Segmentation of leaf veins from microscope images

Requires

• Python 3.6
• pipenv

Setup

Package management and python virtual environments are managed by pipenv

run

pipenv install

or

make setup_environment

To have pipenv install required packages and setup a virtual environment.

Build Training Data sets

To create the training dataset, run:

  make clean_targets
  make dataset

This will create the folder data/processed/veins with the training data enclosed.

The traced (target) images are slightly modified to remove non-binary pixels and to remove small gaps between lines that should be connected.

The real images are ready out of the box. They are just copied from the raw data folder to the processed data folder.

Image names in the data/raw/Veins machine learning with Chris/blacklist.csv will be ignored.

Segment Images

The models require ~12.5Gb of memory to segment an image. Segmentations were performed on the CPU due to the large memory requirements. On the cpu, one images takes ~1.5min to segment.

FINAL

To segment a folder of images using the final trained model, see the src/data/segment_extract_length_folder_sapleo2.sh. This enables segmentation of images on UGA's sapelo2 cluster.

v3_dilations.

v3_dilations was the last model derivation tried before training the last model. It is kept here for posterity.
The model weights from the noAug_dilations_epoch800 save point is used.

To segment the images set aside for testing using the v3_dilations model:

  make models/v3/results/dilations/test/

To segment all images using the v3_dilations model:

  make models/v3/results/dilations/train/

To segment the vein images for other species:

  make models/v3/results/generalization/

The other species vein images were also run through the network at 4x scale by adding image = rescale(image,4,anti_aliasing=True) to segment_images.py at:

...
for image_path in tqdm(images):
    image = io.imread(image_path)
    image = rescale(image,4,anti_aliasing=True)
...

To extract vein lengths (in pixels) from segmentation probabilities:

make v3_results

This will save the measurements to a sqlite database at reports/data/results.sqlite in the tables:

• v3_test_vein_length: For the test image data set
• v3_train_vein_length: For the train image data set

To add manual measurements to the results database

make manual_measurments

This will save the measurements to a sqlite database at reports/data/results.sqlite in a table named manually_measured

To export all data in database to csv files:

make export_csv

Will result in csv files in the reports/csv/ directory.

Notebooks

The notbooks directory contains all jupyter notebooks. Jupyter notebook can be started using:

pipenv run jupyter notebook

• notebooks/V3/ contains the analysis of the many V3 models.
• notebooks/CompareToData.ipynb compares the V3 model segmentations to the manually measured vein lengths.
  • notebooks/Correlation.png is generated from this notebook
• notebooks/Model Generalziation.ipynb is an exploration of how the model generalizes to other vein species images
• notebooks/Segment Whole Images.ipynb was used to explore how much padding would be needed to segment whole images using the neural network.
• notebooks/Clean_Up_Target_Imgaes.ipynb explores cleaning up the segmentation masks originally provided to me by the Burke lab.

Project Organization

Based off of cookiecutter data science:

├── LICENSE
├── Makefile           <- Makefile with commands like `make data` or `make train`
├── README.md          <- The top-level README for developers using this project.
├── data
│   ├── external       <- Data from third party sources.
│   ├── interim        <- Intermediate data that has been transformed.
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
│
├── docs               <- A default Sphinx project; see sphinx-doc.org for details
│
├── models             <- Trained and serialized models, model predictions, or model summaries
│
├── notebooks          <- Jupyter notebooks. Naming convention is a number (for ordering),
│                         the creator's initials, and a short `-` delimited description, e.g.
│                         `1.0-jqp-initial-data-exploration`.
│
├── references         <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
│   └── figures        <- Generated graphics and figures to be used in reporting
│
├── Pipfile            <- The requirements file for reproducing the analysis environment,
│                         generated by pip
│
├── setup.py           <- makes project pip installable (pip install -e .) so src can be imported
├── src                <- Source code for use in this project.
│   ├── __init__.py    <- Makes src a Python module
│   │
│   ├── data           <- Scripts to download or generate data
│   │   └── make_dataset.py
│   │
│   ├── features       <- Scripts to turn raw data into features for modeling
│   │   └── build_features.py
│   │
│   ├── models         <- Scripts to train models and then use trained models to make
│   │   │                 predictions
│   │   ├── predict_model.py
│   │   └── train_model.py
│   │
│   └── visualization  <- Scripts to create exploratory and results oriented visualizations
│       └── visualize.py
│
└── tox.ini            <- tox file with settings for running tox; see tox.testrun.org

---

Project based on the cookiecutter data science project template. #cookiecutterdatascience