This package consists of Kedro pipelines for preprocessing images using TensorFlow. I use it for Dermatology workflows and multimodal machine learning. Use this template that imports this package for multimodal ML. This package includes Kedro datasets for loading weights (as in CheXnet weights to a DenseNet121) and downloading pre-trained models from TF hub.
- The download pipeline downloads online images defined in a csv file for multilabel classification. The labels are added to the filename. The csv format is:
id, url, labels
1, https://somesite.com/someimage.jpg,dog|black|grey
- The folder pipeline creates TensorFlow dataset from a folder of images with labels as subfolders.
- The multilabel pipeline processes files downloaded by the 'download' pipeline and create a dataset with images and labels. The labels are extracted from the filename. Example: _dog_black.jpg
- Add labels in parameters.yml
master_labels: ["cat", "dog", "white", "black", "tan"]
val_size: 0.2
- pip install git+https://github.com/dermatologist/kedro-tf-image.git
from kedro_tf_image.pipelines import preprocess
download = preprocess.create_download_pipeline(
input="csvdata", output="imageset") #input is csv
folder = preprocess.create_folder_pipeline(
input="imagefolder", output="processeddataset")
multilabel = preprocess.create_multilabel_pipeline(input="imageset", output="processeddataset")
# check output keys in the catalog below
imageset:
type: PartitionedDataSet
dataset: {
"type": "kedro_tf_image.extras.datasets.tf_image_dataset.TfImageDataSet",
"imagedim": 224,
"preprocess_input": "tensorflow.keras.applications.resnet50.preprocess_input"
}
path: data/01_raw/imageset
filename_suffix: ".jpg"
csvdata:
type: pandas.CSVDataSet
filepath: data/01_raw/csvfile.csv
imagefolder:
type: kedro_tf_image.extras.datasets.tf_image_folder.TfImageFolder
folderpath: "/path/to/images"
imagedim: 224
load_args:
validation_split: 0.2
seed: 123
batch_size: 1
processeddataset:
type: kedro_tf_image.extras.datasets.tf_image_processed.TfImageProcessed
folderpath: data/02_intermediate/
imagedim: 224
# This is required as copy_mode: assign is needed for TF datasets
datasetinmemory:
type: MemoryDataSet
copy_mode: assign
- kedro_tf_image.extras.datasets.tf_image_dataset.TfImageDataSet - Load single images
- kedro_tf_image.extras.datasets.tf_image_folder.TfImageFolder - Load a folder of images
- kedro_tf_image.extras.datasets.tf_model_weights.TfModelWeights - Read model from weights (Ex: CheXnet with dim 14) (Use create_classification_layer to add a Dense layer of NCLASSES dim)
- kedro_tf_image.extras.datasets.tf_model_download.TfModelDownload - Load model from TF hub.
This is your new Kedro project, which was generated using Kedro 0.17.3.
Take a look at the Kedro documentation to get started.
In order to get the best out of the template:
- Don't remove any lines from the
.gitignorefile we provide - Make sure your results can be reproduced by following a data engineering convention
- Don't commit data to your repository
- Don't commit any credentials or your local configuration to your repository. Keep all your credentials and local configuration in
conf/local/
Declare any dependencies in src/requirements.txt for pip installation and src/environment.yml for conda installation.
To install them, run:
kedro install
You can run your Kedro project with:
kedro run
Have a look at the file src/tests/test_run.py for instructions on how to write your tests. You can run your tests as follows:
kedro test
To configure the coverage threshold, look at the .coveragerc file.
To generate or update the dependency requirements for your project:
kedro build-reqs
This will copy the contents of src/requirements.txt into a new file src/requirements.in which will be used as the source for pip-compile. You can see the output of the resolution by opening src/requirements.txt.
After this, if you'd like to update your project requirements, please update src/requirements.in and re-run kedro build-reqs.
Further information about project dependencies
Note: Using
kedro jupyterorkedro ipythonto run your notebook provides these variables in scope:context,catalog, andstartup_error.
To use Jupyter notebooks in your Kedro project, you need to install Jupyter:
pip install jupyter
After installing Jupyter, you can start a local notebook server:
kedro jupyter notebook
To use JupyterLab, you need to install it:
pip install jupyterlab
You can also start JupyterLab:
kedro jupyter lab
And if you want to run an IPython session:
kedro ipython
You can move notebook code over into a Kedro project structure using a mixture of cell tagging and Kedro CLI commands.
By adding the node tag to a cell and running the command below, the cell's source code will be copied over to a Python file within src/<package_name>/nodes/:
kedro jupyter convert <filepath_to_my_notebook>
Note: The name of the Python file matches the name of the original notebook.
Alternatively, you may want to transform all your notebooks in one go. Run the following command to convert all notebook files found in the project root directory and under any of its sub-folders:
kedro jupyter convert --all
To automatically strip out all output cell contents before committing to git, you can run kedro activate-nbstripout. This will add a hook in .git/config which will run nbstripout before anything is committed to git.
Note: Your output cells will be retained locally.
Further information about building project documentation and packaging your project