Background/Objectives: Chronic wounds of the lower extremities, particularly arterial and venous ulcers, represent a significant and costly challenge in medical care. To assist in differential diagnosis, we aim to evaluate various advanced deep-learning models for classifying arterial and venous ulcers and visualize their decision-making processes. Methods: A retrospective dataset of 607 images (198 arterial and 409 venous ulcers) was used to train five convolutional neural networks: ResNet50, ResNeXt50, ConvNeXt, EfficientNetB2, and EfficientNetV2. Model performance was assessed using accuracy, precision, recall, F1-score, and ROC-AUC. Grad-CAM was applied to visualize image regions contributing to classification decisions. Results: The models demonstrated high classification performance, with accuracy ranging from 72% (ConvNeXt) to 98% (ResNeXt50). Precision and recall values indicated strong discrimination between arterial and venous ulcers, with EfficientNetV2 achieving the highest precision. Conclusions: AI-assisted classification of venous and arterial ulcers offers a valuable method for enhancing diagnostic efficiency.

The analysis of the success of any exhibition depends on the visitor experience. The data required for analysis is usually painstakingly collected by hand. We propose a large-scale optical tracking pipeline to estimate visitor data such as visit trajectory, duration, and, potentially, other personal parameters like age, weight, and sex, yet remain ethically acceptable by obtaining visitor consent. We further show, in preliminary results, that the edge device has a localization error of 0.64 meters and an average precision of 0.2. With this work-in-progress, we intend to ensure a viable alternative to current data collection processes in museum research.Abstract

Understanding human attention mechanisms is crucial for advancing both vision science and artificial intelligence. While numerous computational models of free-viewing have been proposed, less is known about the mechanisms underlying task-driven image exploration. To address this gap, we introduce NevaClip, a novel zero-shot method for predicting visual scanpaths. NevaClip leverages contrastive language-image pretrained (CLIP) models in conjunction with human-inspired neural visual attention (NeVA) algorithms. By aligning the representation of foveated visual stimuli with associated captions, NevaClip uses gradient-driven visual exploration to generate scanpaths that simulate human attention. We also present CapMIT1003, a new dataset comprising captions and click-contingent image explorations collected from participants engaged in a captioning task. Based on the established MIT1003 benchmark, which includes eye-tracking data from free-viewing conditions, CapMIT1003 provides a valuable resource for studying human attention across both free-viewing and task-driven contexts. Additionally, we demonstrate NevaClip’s performance on the publicly available AiR-D dataset, which includes visual question answering (VQA) tasks. Experimental results show that NevaClip outperforms existing unsupervised computational models in scanpath plausibility across captioning, VQA, and free-viewing tasks. Furthermore, we demonstrate that NevaClip’s performance is sensitive to caption accuracy, with misleading captions leading to inaccurate scanpath behaviors. This underscores the importance of caption guidance in attention prediction and highlights NevaClip’s potential to advance our understanding of task-driven human attention mechanisms. Together, NevaClip and CapMIT1003 offer significant contributions to the field, providing new tools for studying and simulating human visual attention.

Urban trees face intensifying drought stress, and soil water storage and its dynamics are important for the drought resilience of urban trees. Existing evidence is based on process-based growth and ecosystem-service models, on water-balance studies of urban green spaces and irrigation, and on single-city monitoring campaigns, while multi-city in-situ observations under a consistent methodology are missing. We present a continuously growing database of 555 urban tree sites in 41 Central European locations, each instrumented with soil moisture sensors and paired with meteorological data, topographic, sun exposure and groundwater descriptors, manually digitised land use around the tree and tree characteristics. An event-based workflow detects rain and irrigation events and extracts two response metrics, the normalised soil moisture increase and the drying time. Random Forest models are built for all combinations of target (soil moisture increase, drying time), event type (rain, irrigation) and measurement depth (30, 60, 90 cm). The models explain up to 69 % of the variance of the moisture increase after irrigation and 40 to 55 % for the other combinations in cross-validation. Grouped cross-validation shows that the rain models transfer to unseen sites within 0.03 in R2 while the irrigation models lose 0.21 to 0.37, therefore encoding a site signature of the few heavily irrigating cities. Rain and irrigation responses differ markedly. The mean air temperature during the event and the pre-event rainfall dominate the rain responses, whereas the pre-event soil moisture and the storage capacity of the substrate control the irrigation responses and how long their effect persists. The unsealed, vegetated area around the tree is the only consistently relevant site characteristic, whereas depth to groundwater shows no effect. The database and workflow are openly available and turn the assessment of planting sites and the verification of irrigation into measurable quantities.