DPP paper is now available online!The
Deep Plant Phenomics
Deep Plant Phenomics (DPP) is a platform for plant phenotyping using deep learning. Think of it as Keras for plant scientists.
Read the doumentation for tutorials, or see the included examples.
DPP is maintained at the Plant Phenotyping and Imaging Research Center (P2IRC) at the University of Saskatchewan.
What's Deep Learning?
Principally, DPP provides deep learning functionality for plant phenotyping and related applications. Deep learning is a category of techniques which encompasses many different types of neural networks. Deep learning techniques lead the state of the art in many image-based tasks, including image classification, object detection and localization, image segmentation, and others.
What Can I Do With This?
This package provides two things:
1. Useful tools made possible using pre-trained neural networks
For example, calling
tools.predict_rosette_leaf_count(my_files) will use a pre-trained convolutional neural network to estimate the number of leaves on each rosette plant.
2. An easy way to train your own models
For example, using a few lines of code you can easily use your data to train a convolutional neural network to rate plants for biotic stress. See the tutorial for how the leaf counting model was built.
- Several pre-made networks for common plant phenotyping tasks.
- Automatic batching and input pipeline.
- Loaders for some popular plant phenotyping datasets.
- Plenty of different loaders for your own data, however it exists.
- Predict classes, values, bounding boxes, or segmentations.
- Support for semantic segmentation with fully convolutional networks.
- Tensorboard integration.
- Easy-to-use API for building new models.
- Several data augmentation options.
- Many ready-to-use neural network layers.
- Easy to deploy your own models as a Python function!
Train a simple regression model to rate plants for abiotic stress:
import deepplantphenomics as dpp model = dpp.DPPModel(debug=True) # 3 channels for colour, 1 channel for greyscale channels = 3 # Setup and hyperparameters model.set_problem_type('regression') model.set_batch_size(64) model.set_image_dimensions(256, 256, channels) model.set_learning_rate(0.0001) model.set_maximum_training_epochs(100) model.set_train_test_split(0.8) # Load dataset of images and ground-truth labels model.load_multiple_labels_from_csv('./data/my_labels.csv') model.load_images_with_ids_from_directory('./data') # Simple convolutional neural network model model.add_input_layer() model.add_convolutional_layer(filter_dimension=[5, 5, channels, 32], stride_length=1, activation_function='relu') model.add_pooling_layer(kernel_size=3, stride_length=2) model.add_convolutional_layer(filter_dimension=[5, 5, 32, 32], stride_length=1, activation_function='relu') model.add_pooling_layer(kernel_size=3, stride_length=2) model.add_fully_connected_layer(output_size=256, activation_function='relu') model.add_output_layer() # Train! model.begin_training()
- The package should work on Python 2.7 or 3.x, but if using 2.7 you need the enum34 package installed.
- Install the following dependencies, following the directions provided according to your platform and requirements:
git clone https://github.com/p2irc/deepplantphenomics.git
python setup.py install
Downloading Pre-trained Networks
The package uses Git Large File Storage (git-lfs) to handle the saved network states included in this repository, as they can sometimes be very large.
If you had git-lfs installed when you installed the packages, then you automatically downloaded the saved networks. If you want to download the states after installing the package, then install git-lfs and run
git lfs fetch and then
git lfs pull.
Contributions are always welcome. If you would like to make a contribution, please fork from the develop branch.
If you are interested in research collaborations or want more information regarding this package, please email
If you have a feature request or bug report, please open a new issue.