arXiv subjects

Arik Mitschang

Publications and source records attributed to Arik Mitschang.

3 recordsLinked to original sources

arXiv2026

TornadoNet: Real-Time Building Damage Detection with Ordinal Supervision

We present TornadoNet, a comprehensive benchmark for automated street-level building damage assessment evaluating how modern real-time object detection architectures and ordinal-aware supervision strategies perform under realistic post-disaster conditions. TornadoNet provides the first controlled benchmark demonstrating how architectural design and loss formulation jointly influence multi-level damage detection from street-view imagery, delivering methodological insights and deployable tools for disaster response. Using 3,333 high-resolution geotagged images and 8,890 annotated building instances from the 2021 Midwest tornado outbreak, we systematically compare CNN-based detectors from the YOLO family against transformer-based models (RT-DETR) for multi-level damage detection. Models are trained under standardized protocols using a five-level damage classification framework based on IN-CORE damage states, validated through expert cross-annotation. Baseline experiments reveal complementary architectural strengths. CNN-based YOLO models achieve highest detection accuracy and throughput, with larger variants reaching 46.05% mAP@0.5 at 66-276 FPS on A100 GPUs. Transformer-based RT-DETR models exhibit stronger ordinal consistency, achieving 88.13% Ordinal Top-1 Accuracy and MAOE of 0.65, indicating more reliable severity grading despite lower baseline mAP. To align supervision with the ordered nature of damage severity, we introduce soft ordinal classification targets and evaluate explicit ordinal-distance penalties. RT-DETR trained with calibrated ordinal supervision achieves 44.70% mAP@0.5, a 4.8 percentage-point improvement, with gains in ordinal metrics (91.15% Ordinal Top-1 Accuracy, MAOE = 0.56). These findings establish that ordinal-aware supervision improves damage severity estimation when aligned with detector architecture. Model & Data: https://github.com/crumeike/TornadoNet

cs.CV↗

arXiv2025

Tidal Debris Candidates from the $\omega$ Centauri Accretion Event and its Role in Building Up the Milky Way Halo

We identify stellar tidal debris from the $\omega$ Centauri ($\omega$ Cen) system among field stars in the APOGEE survey via chemical tagging using a neural network trained on APOGEE observations of the $\omega$ Cen core. We find a total of 463 $\omega$ Cen debris candidates have a probability $P > 0.8$ of sharing common patterns in their chemical abundances across a range of individual elements or element combinations, including [C+N], O, Mg, Al, Si, Ca, Ni, and Fe. Some debris candidates show prograde or retrograde disk-like kinematics, but most show kinematics consistent with the accreted halo, showing high radial actions, $J_{R}$, values. We find that a sample of Gaia-Sausage-Enceladus (GES) members are chemically distinct from the $\omega$ Cen core, suggesting that $\omega$ Cen is associated to an independent merger event shaping the Milky Way halo. However, a connection between GSE and $\omega$ Cen cannot be ruled out. A detailed comparison with $N$-body simulations indicates that the $\omega$ Cen progenitor was a massive dwarf galaxy ($\gtrsim 10^8 M_{\odot}$). The existence of a metal-poor high-$\alpha$ chemically homogeneous halo debris is also reported.

astro-ph.GA↗

arXiv2020

Cost Management on Commercial Cloud Platforms

Commercial cloud platforms are a powerful technology for astronomical research. Despite the benefits of cloud computing -- such as on-demand scalability and reduction of systems management overhead -- confusion over how to manage costs remains, for many, one of the biggest barriers to entry. This confusion is exacerbated by the rapid growth in services offered by commercial providers, by the growth in the number of these providers, and by storage, compute, and I/O metered at separate rates -- all of which can change without notice. As a rule, processing is very cheap, storage is more expensive, and downloading is very expensive. Thus, an application that produces large image data sets for download will be far more expensive than an application that performs extensive processing on a small data set. This Birds of a Feather (BoF) session aimed to quantify the above statement by presenting case studies of the costing of astronomy applications on commercial clouds that covered a range of processing scenarios; these presentations were the basis for discussion by the attendees.

astro-ph.IM↗