Search

Uncertainty Quantification (UQ) is crucial for reliable image segmentation. Yet, while the field sees continual development of novel methods, a lack of agreed-upon benchmarks limits their systematic comparison and evaluation: Current UQ methods are typically tested either on overly simplistic toy datasets or on complex real-world datasets that do not allow to discern true uncertainty. To unify both controllability and complexity, we introduce Arctique, a procedurally generated dataset modeled after histopathological colon images. We chose histopathological images for two reasons: 1) their complexity in terms of intricate object structures and highly variable appearance, which yields challenging segmentation problems, and 2) their broad prevalence for medical diagnosis and respective relevance of high-quality UQ. To generate Arctique, we established a Blender-based framework for 3D scene creation with intrinsic noise manipulation. Arctique contains 50,000 rendered images with precise masks as well as noisy label simulations. We show that by independently controlling the uncertainty in both images and labels, we can effectively study the performance of several commonly used UQ methods. Hence, Arctique serves as a critical resource for benchmarking and advancing UQ techniques and other methodologies in complex, multi-object environments, bridging the gap between realism and controllability. All code is publicly available, allowing re-creation and controlled manipulations of our shipped images as well as creation and rendering of new scenes., Comment: 13 pages, 4 figures

Diverse and realistic floor plan data are essential for the development of useful computer-aided methods in architectural design. Today's large-scale floor plan datasets predominantly feature simple floor plan layouts, typically representing single-apartment dwellings only. To compensate for the mismatch between current datasets and the real world, we develop \textbf{Modified Swiss Dwellings} (MSD) -- the first large-scale floor plan dataset that contains a significant share of layouts of multi-apartment dwellings. MSD features over 5.3K floor plans of medium- to large-scale building complexes, covering over 18.9K distinct apartments. We validate that existing approaches for floor plan generation, while effective in simpler scenarios, cannot yet seamlessly address the challenges posed by MSD. Our benchmark calls for new research in floor plan machine understanding. Code and data are open., Comment: ECCV 2024 (incl. Suppl. Mat.)

Heatmaps generated on inputs of image classification networks via explainable AI methods like Grad-CAM and LRP have been observed to resemble segmentations of input images in many cases. Consequently, heatmaps have also been leveraged for achieving weakly supervised segmentation with image-level supervision. On the other hand, losses can be imposed on differentiable heatmaps, which has been shown to serve for (1)~improving heatmaps to be more human-interpretable, (2)~regularization of networks towards better generalization, (3)~training diverse ensembles of networks, and (4)~for explicitly ignoring confounding input features. Due to the latter use case, the paradigm of imposing losses on heatmaps is often referred to as "Right for the right reasons". We unify these two lines of research by investigating semi-supervised segmentation as a novel use case for the Right for the Right Reasons paradigm. First, we show formal parallels between differentiable heatmap architectures and standard encoder-decoder architectures for image segmentation. Second, we show that such differentiable heatmap architectures yield competitive results when trained with standard segmentation losses. Third, we show that such architectures allow for training with weak supervision in the form of image-level labels and small numbers of pixel-level labels, outperforming comparable encoder-decoder models. Code is available at \url{https://github.com/Kainmueller-Lab/TW-autoencoder}.

We introduce a double framing construction for moduli spaces of quiver representations. It allows us to reduce certain sheaf cohomology computations involving the universal representation, to computations involving line bundles, making them amenable to methods from geometric invariant theory. We will use this to show that in many good situations the vector fields on the moduli space are isomorphic as a vector space to the first Hochschild cohomology of the path algebra. We also show that considering the universal representation as a Fourier-Mukai kernel in the appropriate sense gives an admissible embedding of derived categories., Comment: v2: added references, 23 pages, all comments welcome

We explain how Teleman quantization can be applied to moduli spaces of quiver representations to compute the higher cohomology of the endomorphism bundle of the universal bundle. We use this to prove Schofield's partial tilting conjecture, and to show that moduli spaces of quiver representations are (infinitesimally) rigid as varieties., Comment: v2: added references, 26 pages, comments welcome

We describe the point class and Todd class in the Chow ring of a quiver moduli space, building on a result of Ellingsrud-Str{\o}mme. This, together with the presentation of the Chow ring by the second author, makes it possible to compute integrals on quiver moduli. To do so we construct a canonical morphism of universal representations in great generality, and along the way point out its relation to the Kodaira-Spencer morphism. We illustrate the results by computing some invariants of some "small" Kronecker moduli spaces. We also prove that the first non-trivial (6-dimensional) Kronecker quiver moduli space is isomorphic to the zero locus of a general section of $\mathcal{Q}^\vee(1)$ on $\operatorname{Gr}(2,8)$., Comment: 18 pages, now proves the conjecture from v1 and includes data on more Kronecker moduli

Fully convolutional recurrent neural networks (FCRNs) have shown state-of-the-art performance in single-channel speech enhancement. However, the number of parameters and the FLOPs/second of the original FCRN are restrictively high. A further important class of efficient networks is the CRUSE topology, serving as reference in our work. By applying a number of topological changes at once, we propose both an efficient FCRN (FCRN15), and a new family of efficient convolutional recurrent neural networks (EffCRN23, EffCRN23lite). We show that our FCRN15 (875K parameters) and EffCRN23lite (396K) outperform the already efficient CRUSE5 (85M) and CRUSE4 (7.2M) networks, respectively, w.r.t. PESQ, DNSMOS and DeltaSNR, while requiring about 94% less parameters and about 20% less #FLOPs/frame. Thereby, according to these metrics, the FCRN/EffCRN class of networks provides new best-in-class network topologies for speech enhancement., Comment: 5 pages, 5 figures, accepted for Interspeech 2023

In Human-Computer-Interaction, vibrotactile haptic feedback offers the advantage of being independent of any visual perception of the environment. Most importantly, the user's field of view is not obscured by user interface elements, and the visual sense is not unnecessarily strained. This is especially advantageous when the visual channel is already busy, or the visual sense is limited. We developed three design variants based on different vibrotactile illusions to communicate 3D directional cues. In particular, we explored two variants based on the vibrotactile illusion of the cutaneous rabbit and one based on apparent vibrotactile motion. To communicate gradient information, we combined these with pulse-based and intensity-based mapping. A subsequent study showed that the pulse-based variants based on the vibrotactile illusion of the cutaneous rabbit are suitable for communicating both directional and gradient characteristics. The results further show that a representation of 3D directions via vibrations can be effective and beneficial., Comment: CHI EA '23, April 23-28, 2023, Hamburg, Germany

One of the most important magnetic spin structure is the topologically stabilised skyrmion quasi-particle. Its interesting physical properties make them candidates for memory and efficient neuromorphic computation schemes. For the device operation, detection of the position, shape, and size of skyrmions is required and magnetic imaging is typically employed. A frequently used technique is magneto-optical Kerr microscopy where depending on the samples material composition, temperature, material growing procedures, etc., the measurements suffer from noise, low-contrast, intensity gradients, or other optical artifacts. Conventional image analysis packages require manual treatment, and a more automatic solution is required. We report a convolutional neural network specifically designed for segmentation problems to detect the position and shape of skyrmions in our measurements. The network is tuned using selected techniques to optimize predictions and in particular the number of detected classes is found to govern the performance. The results of this study shows that a well-trained network is a viable method of automating data pre-processing in magnetic microscopy. The approach is easily extendable to other spin structures and other magnetic imaging methods.

The blast furnace (BF) is the fundamental tool used in the iron manufacture. Due to the difficulty of accessing direct measurements of the inner phenomena, we determined the density distribution of its internal volume in order to improve its productivity using muography. This is an imaging technique based on the differential absorption of a flux of incident particles, muons, by the target under study, similar to clinical X-ray imaging. Muons are elementary particles that have the property of passing through dense materials, up to hundreds of meters away. Their relative absorption and deviation allows the generation of density distribution images of an object by tracking the number of muons received by a detector, before and after passing through a structure. The incident direction of the detected muons is reconstructed by means of a detector composed of 3 scintillator panels that we moved on 3 positions around the BF. With this technique, we obtained the first 3D image of the internal structure of a BF using a Markov Chain Monte Carlo (MCMC) inverse problem solving algorithm on muon flux data. We were also able to perform a density monitoring of the BF and some of its operating parameters. We distinguished the position and shape of the cohesive zone, a key element in the productivity of a furnace, validating this innovative measurement concept in the application to a BF and opening the field to a series of future experiments to gain both spatial and temporal resolution.

We report on the first observation of photoassociation near the ${}^2 S _\frac 1 2 + {}^3P_1$-asymptote in a mixture of ${}^{87}$Rb and ${}^{170}$Yb. In a search spanning binding energies between 0.1 GHz and 11 GHz, a single pair of interspecies PA resonances is detected around 3.1 GHz. These resonances are characterized by extracting PA rates, binding energies and Zeeman shift coefficients. Using one of these resonances, 2-photon-photoassociaton is performed, improving on previous measurements of the binding energies of the two least bound states in the electronic ground state and demonstrating intercombination line photoassociation as a powerful spectroscopic tool. We discuss implications for pathways towards RbYb molecules in the absolute ground state., Comment: 8 pages, 4 figures

We show that a semi-stable moduli space of representations of an acyclic quiver can be identified with two GIT quotients by reductive groups. One of a quiver Grassmannian of a projective representation, the other of a quiver Grassmannian of an injective representation. This recovers as special cases the classical Gelfand-MacPherson correspondence and its generalization by Hu and Kim to bipartite quivers, as well as the Zelevinsky map for a quiver of Dynkin type A with the linear orientation., Comment: 28 pages, comments are welcome

Nowadays, robots are found in a growing number of areas where they collaborate closely with humans. Enabled by lightweight materials and safety sensors, these cobots are gaining increasing popularity in domestic care, supporting people with physical impairments in their everyday lives. However, when cobots perform actions autonomously, it remains challenging for human collaborators to understand and predict their behavior, which is crucial for achieving trust and user acceptance. One significant aspect of predicting cobot behavior is understanding their motion intention and comprehending how they "think" about their actions. Moreover, other information sources often occupy human visual and audio modalities, rendering them frequently unsuitable for transmitting such information. We work on a solution that communicates cobot intention via haptic feedback to tackle this challenge. In our concept, we map planned motions of the cobot to different haptic patterns to extend the visual intention feedback., Comment: Final CHI LBW 2023 submission: https://dx.doi.org/10.1145/3544549.3585601

We give a moduli-theoretic treatment of the existence and properties of moduli spaces of semistable quiver representations, avoiding methods from geometric invariant theory. Using the existence criteria of Alper--Halpern-Leistner--Heinloth, we show that for many stability functions, the stack of semistable representations admits an adequate moduli space, and prove that this moduli space is proper over the moduli space of semisimple representations. We construct a natural determinantal line bundle that descends to a semiample line bundle on the moduli space and provide new effective bounds for global generation. For an acyclic quiver, we show that this line bundle is ample, thus giving a modern proof of the fact that the moduli space is projective., Comment: 49 pages; comments welcome

Decisions about sharing personal information are not trivial, since there are many legitimate and important purposes for such data collection, but often the collected data can reveal sensitive information about individuals. Privacy-preserving technologies, such as differential privacy (DP), can be employed to protect the privacy of individuals and, furthermore, provide mathematically sound guarantees on the maximum privacy risk. However, they can only support informed privacy decisions, if individuals understand the provided privacy guarantees. This article proposes a novel approach for communicating privacy guarantees to support individuals in their privacy decisions when sharing data. For this, we adopt risk communication formats from the medical domain in conjunction with a model for privacy guarantees of DP to create quantitative privacy risk notifications. We conducted a crowd-sourced study with 343 participants to evaluate how well our notifications conveyed the privacy risk information and how confident participants were about their own understanding of the privacy risk. Our findings suggest that these new notifications can communicate the objective information similarly well to currently used qualitative notifications, but left individuals less confident in their understanding. We also discovered that several of our notifications and the currently used qualitative notification disadvantage individuals with low numeracy: these individuals appear overconfident compared to their actual understanding of the associated privacy risks and are, therefore, less likely to seek the needed additional information before an informed decision. The promising results allow for multiple directions in future research, for example, adding visual aids or tailoring privacy risk communication to characteristics of the individuals., Comment: Accepted to ACM CCS 2022. arXiv admin note: substantial text overlap with arXiv:2204.04061