Data Augmentation is quintessential for training modern AI-based image analysis methods. Through careful modification of the available training data, the training distribution is extended and/or covered in more detail, reducing the potential for overfitting. Batchgenerators is deep-learning framework agnostic and can be integrated into your PyTorch and Tensorflow workloads and …

challengeR is a tool for analyzing and visualizing challenge results in the field of image analysis and beyond.

Challenges have become the de facto standard for benchmarking image analysis algorithms. While the number of challenges is steadily increasing, surprisingly little effort has been invested in ensuring high quality design, execution …

Generalized-YOLOv5 is a modified version of YOLOv5. YOLOv5 itself is a realtime object detection framework designed for natural images. Our Generalized-YOLOv5 version has two crucial extensions. First, an extension that generalizes YOLOv5 also to non-natural images, which enables the usage of state-of-the-art realtime object detection to many domains (e.g. the …

This solution contains a framework for training deep learning-based classification and regression models with Pytorch Lightning. It comprises several architectures, regularization, augmentation and training techniques and aims to provide an easy-to-use suite for experimenting and fast model development using your own datasets. Additionally, extensive experiments on the CIFAR10 dataset provide …

This repository contains an easy-to-use and flexibly customizable framework for developing and training semantic segmentation models. The focus was put on being usable out-of-the-box, without being a black box and giving the possibility to be adapted to individual projects. Therefore several features like pre-trained state-of-the-art models are already provided for …

CrystFEL is a suite of programs for processing serial crystallography data. CrystFEL comprises programs for indexing and integrating diffraction patterns, scaling and merging intensities, simulating patterns, calculating figures of merit for the data and visualising the results.

The Dislocation Ontology Suite (DISOS) is an ontology suite comprising several modules describing materials scientific concepts, representations of dislocations, and different simulation models in the dislocation domain. Currently, DISOS comprises three ontologies, including:
- Dislocation Ontology (DISO)
- Crystal Structure Ontology (CSO)
- Crystallographic Defect Ontology (CDO)

This repository contains a framework for benchmarking failure detection methods under distribution shifts in image classification featuring implementations of baseline methods, data sets and metrics. Additionally to a wide range of built-in methods, this framework can be used to evaluate and compare new methods, models and datasets.

This repository contains a framework which allows for enhanced model interpretability and model robustness. The framework utilizes interpretable disentangled representations for downstream-task prediction which are in a second step visualized and evaluated through a multipath-attribution mapping.

Simultaneous localisation and categorisation of objects, also referred to as object detection, is of high relevance for many applications which rely on rating of objects rather than e.g. pixels or entire images. For this task, the cumbersome and iterative process of method configuration constitutes a major research and application bottleneck. …

When confronted with a new dataset, the application of AI-based segmentation methods is non-trivial. Substantial manual effort is required to optimize the network topology, preprocessing, training procedure and inference strategy to the task at hand. Such adaptations are non-trivial and require substantial expertise to ensure optimal performance on each dataset. …

This python based tool extracts metadata from machine generated scanning microscopy images in the TIFF format. Currently only metadata from SEMs manufactured by Zeiss, and acquired by their own software is supported. In future, metadata from machines developed by other manufacturers, and produced by other softwares would also be included. …