First versions of the two libraries developed within the DeepHealth project, EDDL (European Distributed Deep Learning Library) and ECVL (European Computer Vision Library) are publicly available in GitHub.
EDDL and ECVL are two of the major outcomes of the DeepHealth project, forming together with the associated front-end and backend application, the DeepHealth toolkit.
- EDDL is an optimized tensor library for distributed deep learning with hardware transparency support for CPUs, GPUs and FPGAs. It is a general purpose software that can be used to design, train, evaluate and put in production any network topology designed by a Deep Learning expert.
- ECVL is a library that facilitates the integration and exchange of data between existing Computer Vision and image processing libraries. The library also provides high-level Computer Vision functionalities thanks to specialized/accelerated versions of some CV algorithms commonly used in combination with Deep Learning (DL) algorithms, ready to be used with EDDL.
Usable versions of their C++ and Python versions are already available, together with installation instructions and examples of use. All information and documentation in the GitHub repository of the EDDL and ECVL.
Development is still on-going, so, more updates are yet to come, in special towards their distributed versions and their adaptation to be efficiently executed on HPC and Cloud infrastructures.
At the same time, the research group led by María de la Iglesia-Vayá of the Foundation for the Promotion of Health and Biomedical Research in the Valencian Community (FISABIO) in the DeepHealth Consortia, has published a large annotated open dataset of RX and CT images of COVID19 patients.
The BIMCV COVID-19+ dataset is a large open multi-institutional databank that provides the open scientific community with data of clinical-scientific value that will help the early detection and evolution of COVID-19. Making the information accessible to the scientific community worldwide will undoubtedly maximize the usefulness of the data and its exploitation for training predictive models, and it is already in use within the DeepHealth Project to test and validate DeepHealth libraries.
This first iteration of the database includes 1,380 chest X-ray images CXR, 885 images DX and 163 computed tomography studies from 1,311 COVID-19+ patients. This is, to the best of our knowledge, the largest COVID-19+ dataset of images available in an open format, specifically Medical Imaging Data Structure (MIDS); more iterations will be made available to the scientific community shortly.