The DPP paper is now available online!
Deep Plant Phenomics (DPP) is a platform for plant phenotyping using deep learning. Think of it as Keras for plant scientists.
DPP integrates Tensorflow for learning and PlantCV for image processing. This means that it is able to run on both CPUs and GPUs, and scale easily across devices.
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. 🌾🇨🇦
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.
This package provides two things:
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.
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.
- Automatic segmentation using PlantCV and bounding box regression.
- Multi-threading for pre-processing tasks.
- Tensorboard integration.
- Easy-to-use API for building new models.
- Support for both classification and regression problems.
- Data augmentation options.
- Many ready-to-use neural network layers.
Train a simple model to classify species:
import deepplantphenomics as dpp
model = dpp.DPPModel(debug=True)
# 3 channels for colour, 1 channel for greyscale
channels = 3
# Setup and hyperparameters
model.set_batch_size(128)
model.set_image_dimensions(256, 256, channels)
model.set_learning_rate(0.001)
model.set_maximum_training_epochs(700)
model.set_train_test_split(0.75)
# Load dataset
model.load_dataset_from_directory_with_auto_labels('./data')
# Specify pre-processing steps
model.add_preprocessing_step('auto-segmentation')
# 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_convolutional_layer(filter_dimension=[5, 5, 32, 64], 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:
- Tensorflow (1.0 or later)
- PlantCV (Only required for the
auto-segmentation
preprocessor)
git clone https://github.com/p2irc/deepplantphenomics.git
python setup.py install
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 [email protected]
.
If you have a feature request or bug report, please open a new issue.