Constanze Schwan

Pretrained cell segmentation models have simplified and accelerated microscopy image analysis, but they often perform poorly on new and challenging datasets. Although these models can be adapted to new datasets with only a few annotated images, the effectiveness of fine-tuning depends critically on which images are selected for annotation. To address this, we propose CGMD (Centrality-Guided Maximum Diversity), a novel algorithm that identifies a small set of images that are maximally diverse with respect to each other in the pretrained feature space. We evaluate CGMD under an extremely low annotation budget of just two images per dataset for fine-tuning the pretrained Cellpose Cyto2 model on four different 2D+t datasets from the Cell Tracking Challenge. CGMD consistently outperforms six competitive active learning and subset selection methods and approaches the performance of fully supervised fine-tuning. The results show that centrality-guided maximum diversity subset selection enables stable and annotation-efficient fine-tuning of pretrained cell segmentation models. The code is publicly available at: https://github.com/eiram-mahera/cgmd .

Humans learn movements naturally, but it takes a lot of time and training to achieve expert performance in motor skills. In this review, we show how modern technologies can support people in learning new motor skills. First, we introduce important concepts in motor control, motor learning and motor skill learning. We also give an overview about the rapid expansion of machine learning algorithms and sensor technologies for human motion analysis. The integration between motor learning principles, machine learning algorithms and recent sensor technologies has the potential to develop AI-guided assistance systems for motor skill training. We give our perspective on this integration of different fields to transition from motor learning research in laboratory settings to real world environments and real world motor tasks and propose a stepwise approach to facilitate this transition.

State-of-the-art methods in image-based robotic grasping use deep convolutional neural networks to determine the robot parameters that maximize the probability of a stable grasp given an image of an object. Despite the high accuracy of these models they are not applied in industrial order picking tasks to date. One of the reasons is the fact that the generation of the training data for these models is expensive. Even though this could be solved by using a physics simulation for training data generation, another even more important reason is that the features that lead to the prediction made by the model are not human-readable. This lack of interpretability is the crucial factor why deep networks are not found in critical industrial applications. In this study we suggest to reformulate the task of robotic grasping as three tasks that are easy to assess from human experience. For each of the three steps we discuss the accuracy and interpretability. We outline how the proposed three-step model can be extended to depth images. Furthermore we discuss how interpretable machine learning models can be chosen for the three steps in order to be applied in a real-world industrial environment.