Publicly available software will include tools for the homogenization of highly heterogeneous images (e.g. compound images), for weakly supervised extraction of multimodal semantic concepts from text and images and to visualize, navigate and refine knowledge structures.


MedTAG is an open-source biomedical annotation tool for diagnostic reports. MedTAG provides effective and intuitive tools to tag biomedical concepts contained in clinical reports.


Processing Megapixel Images

PyTorch implementation of the paper: “Processing Megapixel Images with Deep Attention-Sampling Models”.

Compound Figure Separator

Complete pipeline for compound figures separation and association to related text (ExaMode Deliverable 3.2).


NanoWeb, a Web-oriented search system that allows to search, access, and explore nanopublications on the Web.

ExaMode CERT

ExaMode CERT allows the user to extract both the entities and concepts from user-provided Colon Cancer-related medical reports.


Multi-resolution convolutional neural networks for semantic segmentation in histopathology whole-slide images.

You can try the algorithm here.

Few-Shot Detection

We developed a deep learning system for weakly supervised object detection in digital pathology whole slide images. We designed the system to be organ- and object-agnostic, and to be adapted on-the-fly to detect novel objects based on a few examples provided by the user.


This repository is developed as part of the Examode EU project, and is meant for conducting experiments for large field-of-view semantic segmentation. The current codebase supports CAMELYON16 and CAMELYON17, and supports efficient execution on multi-node CPU clusters, as well as multi-node, multi-GPU clusters. Models using very large FoV (> 1024×1024) can be trained on multi-GPU cluster, using the instructions below. The models adapted for the use case of semantic segmentation of malignant tumor regions are:


Flow-based generative models parameterize probability distributions through an invertible transformation and can be trained by maximum likelihood. Invertible residual networks provide a flexible family of transformations where only Lipschitz conditions rather than strict architectural constraints are needed for enforcing invertibility.