homeHome exploreGetting Started codeDevelopment peopleAbout bookDocumentation
ELEKTRONN Logo

ELEKTRONN

ELEKTRONN is a deep learning toolkit that makes powerful neural networks accessible to scientists outside of the machine learning community.

Our focus lies on high troughput analysis of large scale 2D and 3D images with convolutional neural networks (CNNs).

  • book Documentation
  • get_app Install (v1.0.9)
  • code GitHub Repository
About Machine Learning

Many data analysis and classification tasks can be formulated as a machine learning problem: without explicit expert knowledge or manual guidance artificial neural networks can be trained to map certain inputs (e.g. raw images) to outputs (e.g. probability maps for classes). This mapping is found by training on a set of exemplary “input-output pairs” — training examples — that must be provided initially. Once a good mapping has been found, it can be used to make predictions on new data.

Membrane and mitochondria probability maps. Predicted with a CNN with recursive training.

Data: zebra finch area X dataset j0126 by Jörgen Kornfeld.

Flexible Training

ELEKTRONN can be used for machine learning tasks formulated on image data or flat feature vectors. We provide a ready-made, flexible training pipeline with helpful utility functions and a comprehensive documentation. See our list of features for more details.

Easy Usage & Optimized Runtimes

ELEKTRONN is written in Python and based on the Theano framework. It can be accelerated with NVIDIA GPUs and delivers state-of-the art performance, including fast training and inference on massive image datasets.

Open Source

ELEKTRONN is a scientific open source project and licensed under the GPL. The code is available at our GitHub repository or can be installed with conda or pip.