Search

Snow degradations present formidable challenges to the advancement of computer vision tasks by the undesirable corruption in outdoor scenarios. While current deep learning-based desnowing approaches achieve success on synthetic benchmark datasets, they struggle to restore out-of-distribution real-world snowy videos due to the deficiency of paired real-world training data. To address this bottleneck, we devise a new paradigm for video desnowing in a semi-supervised spirit to involve unlabeled real data for the generalizable snow removal. Specifically, we construct a real-world dataset with 85 snowy videos, and then present a Semi-supervised Video Desnowing Network (SemiVDN) equipped by a novel Distribution-driven Contrastive Regularization. The elaborated contrastive regularization mitigates the distribution gap between the synthetic and real data, and consequently maintains the desired snow-invariant background details. Furthermore, based on the atmospheric scattering model, we introduce a Prior-guided Temporal Decoupling Experts module to decompose the physical components that make up a snowy video in a frame-correlated manner. We evaluate our SemiVDN on benchmark datasets and the collected real snowy data. The experimental results demonstrate the superiority of our approach against state-of-the-art image- and video-level desnowing methods.

Identifying client needs to provide optimal services is crucial in tourist destination management. The events held in tourist destinations may help to meet those needs and thus contribute to tourist satisfaction. As with product management, the creation of hierarchical catalogs to classify those events can aid event management. The events that can be found on the internet are listed in dispersed, heterogeneous sources, which makes direct classification a difficult, time-consuming task. The main aim of this work is to create a novel process for automatically classifying an eclectic variety of tourist events using a hierarchical taxonomy, which can be applied to support tourist destination management. Leveraging data science methods such as CRISP-DM, supervised machine learning, and natural language processing techniques, the automatic classification process proposed here allows the creation of a normalized catalog across very different geographical regions. Therefore, we can build catalogs with consistent filters, allowing users to find events regardless of the event categories assigned at source, if any. This is very valuable for companies that offer this kind of information across multiple regions, such as airlines, travel agencies or hotel chains. Ultimately, this tool has the potential to revolutionize the way companies and end users interact with tourist events information.

Partial differential equations (PDEs) are widely used to model complex physical systems, but solving them efficiently remains a significant challenge. Recently, Transformers have emerged as the preferred architecture for PDEs due to their ability to capture intricate dependencies. However, they struggle with representing continuous dynamics and long-range interactions. To overcome these limitations, we introduce the Mamba Neural Operator (MNO), a novel framework that enhances neural operator-based techniques for solving PDEs. MNO establishes a formal theoretical connection between structured state-space models (SSMs) and neural operators, offering a unified structure that can adapt to diverse architectures, including Transformer-based models. By leveraging the structured design of SSMs, MNO captures long-range dependencies and continuous dynamics more effectively than traditional Transformers. Through extensive analysis, we show that MNO significantly boosts the expressive power and accuracy of neural operators, making it not just a complement but a superior framework for PDE-related tasks, bridging the gap between efficient representation and accurate solution approximation.

Particle-producing targets in high-energy research facilities are often made from refractory metals, and they typically require dedicated cooling systems due to the challenging thermomechanical conditions they experience. However, direct contact of water with target blocks can induce erosion, corrosion, and embrittlement, especially of tungsten (W). One approach to overcoming this problem is cladding the blocks with tantalum (Ta). Unfortunately, Ta generates high decay heat when irradiated, raising safety concerns in the event of a loss-of-cooling accident. This study explored the capacity of niobium (Nb) and its alloys to form diffusion bonds with W and TZM (a molybdenum alloy with titanium and zirconium). This is because the Beam Dump Facility (BDF), a planned new fixed-target installation in CERN's North Area, uses these target materials. The bonding quality of pure Nb, Nb1Zr, and C103 (a Nb alloy with 10% hafnium and 1% titanium) with TZM and W obtained using hot isostatic pressing (HIP) was evaluated. The effects of different HIP temperatures and the introduction of a Ta interlayer were examined. Optical microscopy indicated promising bonding interfaces, which were further characterized using tensile tests and thermal-diffusivity measurements. Their performance under high-energy beam impact was validated using thermomechanical simulations. C103 exhibited higher interface strengths and safety factors than Ta2.5W, positioning it as a potential alternative cladding material for the BDF production target. The findings highlight the viability of Nb-based materials, particularly C103, for improving operational safety and efficiency in fixed-target physics experiments; however, considerations regarding the long half-life of 94Nb require further attention.

The Beam Dump Facility (BDF) is a planned fixed-target installation in CERN's North Area, set to start operating in 2031. A proton beam of 400 GeV/c will be delivered in 1 s pulses of 4e13 protons every 7.2 s, amounting to 4e19 protons on target (PoT) per year, with an average thermal power deposition of 305 kW. The experiment requires high-Z and high-density materials and involves challenging thermomechanical conditions; hence, a water-cooled refractory metal core is proposed. To prevent corrosion and erosion, the target materials - tungsten and a molybdenum-based alloy, TZM-, are clad with a tantalum alloy using hot isostatic pressing. To verify the reliability of the target design and manufacturing process, a reduced-scale prototype target was manufactured and irradiated with 2.4e16 PoT in 2018. This paper presents the findings of the post-irradiation examination (PIE) conducted 2.5 years later. The PIE consisted of non-destructive and destructive activities, including film imaging, microscopy, metrology, ultrasonic testing, microstructural analysis, and mechanical and thermal characterization. No irradiation effects resulting from thermal stresses, such as geometrical changes or cracks in the cladding, were observed, and it was determined that the bonding between the core and cladding materials was robust. Moreover, no changes were detected in the microstructures or mechanical properties of the bulk materials. In conclusion, the robustness of the target baseline design and manufacturing process was confirmed, validating its suitability for operating under the desired conditions. Nonetheless, the observed brittleness of sintered tungsten indicates a potential area for improvement to enhance the lifetime of the BDF target. Overall, the examinations provided valuable insights into the performance of the prototype and indicated potential refinements for the future BDF target complex.

Purpose: To provide real-time quantitative organ-specific information - specifically placental and brain T2* - to allow optimization of the MR examination to the individual patient. Methods: A FIRE-based real-time setup segmenting placenta and fetal brain in real-time, performing T2* fitting and analysis and calculation of the centile was implemented. A nn-UNet were trained and tested on 2989 datasets for the fetal brain and a second one trained on 210 datasets for the placenta for automatic segmentation. T2* normal curves were obtained from 106 cases and prospective evaluation was performed on 10 cases between 35 and 39 weeks GA. Results: Quantitative brain and placental T2* maps and centiles were available in all prospective cases within 30 seconds. The robustness of the method was shown with intra-scan repeats (mean difference 1.04+/-12.39 ms for fetal brain and -3.15+/-8.88 ms for placenta) and direct validation with vendor-processed offline results (mean difference 1.62+/-4.33 ms for fetal brain and 0.16+/-6.19 ms for placenta). Discussion and Conclusion: Real-time available organ-specific quantitative information enables more personalized MR examinations, selection of the most pertinent sequences and thus the promise of reduced recalls and specific insights into tissue properties. The here enabled placental T2*, demonstrated in multiple recent studies to be a biomarker sensitive to a range of pregnancy complications, and fetal brain T2* will be explored in further studies in pregnancies with pre-eclampsia, growth restriction as a way of enabling future MR-guided fetal interventions.

Since the discovery of the accelerated expansion of the Universe in 1998, modified gravity (MG) theories have attracted considerable attention as alternatives to dark energy (DE). While distinguishing the effects of MG from those of DE using cosmic background expansion alone is difficult, the large-scale structure is expected to differ significantly. Among the plethora of MG models, we are particularly interested in those that introduce a scale dependence in the growth of perturbations; specifically, theories that introduce fifth forces mediated by scalar fields with a finite range accessible to cosmological probes. This is the case with $f(R)$ gravity, which is widely regarded as the most studied model in cosmology. In this work, we utilize, for the first time, the full-shape power spectrum of galaxies to constrain scale-dependent modified gravity theories. By using BOSS DR12 dataset, along with a BBN prior on $ \omega_b $ and a Planck 2018 prior on $ n_s $, we obtain an upper bound of $ |f_{R0}| < 5.89 \times 10^{-6} $ at 68% confidence level (c.l.) and $ < 1.53 \times 10^{-5} $ at 95% c.l. for the Hu-Sawicki ($ n=1 $) model. We discuss that it is highly unlikely these constraints will be significantly improved by future galaxy spectroscopic catalogs, such as DESI and Euclid., Comment: 7 pages, 3 figures

We explore the Schmidt-Kennicutt (SK) relations and the star formation efficiency per free-fall time ($\eff$), mirroring observational studies, in numerical simulations of filamentary molecular clouds undergoing gravitational contraction. We find that {\it a)} collapsing clouds accurately replicate the observed SK relations for galactic clouds and {\it b)} the so-called efficiency per free-fall time ($\eff$) is small and constant in space and in time, with values similar to those found in local clouds. This constancy is a consequence of the similar radial scaling of the free-fall time and the internal mass in density structures with spherically-averaged density profiles near $r^{-2}$. We additionally show that {\it c)} the star formation rate (SFR) increases rapidly in time; {\it d)} the low values of $\eff$ are due to the different time periods over which $\tauff$ and $\tausf$ are evaluated, together with the fast increasing SFR, and {\it e)} the fact that star clusters are significantly denser than the gas clumps from which they form is a natural consequence of the fast increasing SFR, the continuous replenishment of the star-forming gas by the accretion flow, and the near $r^{-2}$ density profile generated by the collapse Finally, we argue that the interpretation of $\eff$ as an efficiency is problematic because its maximum value is not bounded by unity, and because the total gas mass in the clouds is not fixed, but rather depends on the environment where clouds are embedded. In summary, our results show that the SK relation, the typical observed values of $\eff$, and the mass density of clusters arise as a natural consequence of gravitational contraction., Comment: Submitted to MNRAS

Reaction-Diffusion systems arise in diverse areas of science and engineering. Due to the peculiar characteristics of such equations, analytic solutions are usually not available and numerical methods are the main tools for approximating the solutions. In the last decade, artificial neural networks have become an active area of development for solving partial differential equations. However, several challenges remain unresolved with these methods when applied to reaction-diffusion equations. In this work, we focus on two main problems. The implementation of homogeneous Neumann boundary conditions and long-time integrations. For the homogeneous Neumann boundary conditions, we explore four different neural network methods based on the PINN approach. For the long time integration in Reaction-Diffusion systems, we propose a domain splitting method in time and provide detailed comparisons between different implementations of no-flux boundary conditions. We show that the domain splitting method is crucial in the neural network approach, for long time integration in Reaction-Diffusion systems. We demonstrate numerically that domain splitting is essential for avoiding local minima, and the use of different boundary conditions further enhances the splitting technique by improving numerical approximations. To validate the proposed methods, we provide numerical examples for the Diffusion, the Bistable and the Barkley equations and provide a detailed discussion and comparisons of the proposed methods., Comment: 35 pages, 12 figures, research paper

This paper proposes an algorithm for synthesis of clock-follow-data designs that provides robustness against timing violations for RSFQ circuits while maintaining high performance and minimizing area costs. Since superconducting logic gates must be clocked, managing data flow is a challenging problem that often requires the insertion of many path balancing D Flips (DFFs) to properly sequence data, leading to a substantial increase in area. To address this challenge, we present an algorithm to insert DFFs into clock-follow-data RSFQ circuits that partially balances the delays within the circuit to achieve a target throughput while minimizing area. Our algorithm can account for expected timing variations and, by adjusting the bias of the clock network and clock frequency, we can mitigate unexpected timing violations post-fabrication. Quantifying the benefits of our approach with a benchmark suite with nominal delays, our designs offer an average 1.48x improvement in area delay product (ADP) over high frequency full path balancing (FPB) designs and a 2.07x improvement in ADP over the state of the art robust circuits provided by state-of-the-art (SOTA) multi-phase clocking solutions.

Imaging tasks are typically tackled using a structured optimization framework. This paper delves into a class of algorithms for difference-of-convex (DC) structured optimization, focusing on minimizing a DC function along with a possibly nonconvex function. Existing DC algorithm (DCA) versions often fail to effectively handle nonconvex functions or exhibit slow convergence rates. We propose a novel inertial proximal DC algorithm in Bregman geometry, named iBPDCA, designed to address nonconvex terms and enhance convergence speed through inertial techniques. We provide a detailed theoretical analysis, establishing both subsequential and global convergence of iBPDCA via the Kurdyka-{\L}ojasiewicz property. Additionally, we introduce a Plug-and-Play variant, PnP-iBPDCA, which employs a deep neural network-based prior for greater flexibility and robustness while ensuring theoretical convergence. We also establish that the Gaussian gradient step denoiser used in our method is equivalent to evaluating the Bregman proximal operator for an implicitly weakly convex functional. We extensively validate our method on Rician noise and phase retrieval. We demonstrate that iBPDCA surpasses existing state-of-the-art methods.

Sparse-View Computed Tomography (SVCT) offers low-dose and fast imaging but suffers from severe artifacts. Optimizing the sampling strategy is an essential approach to improving the imaging quality of SVCT. However, current methods typically optimize a universal sampling strategy for all types of scans, overlooking the fact that the optimal strategy may vary depending on the specific scanning task, whether it involves particular body scans (e.g., chest CT scans) or downstream clinical applications (e.g., disease diagnosis). The optimal strategy for one scanning task may not perform as well when applied to other tasks. To address this problem, we propose a deep learning framework that learns task-specific sampling strategies with a multi-task approach to train a unified reconstruction network while tailoring optimal sampling strategies for each individual task. Thus, a task-specific sampling strategy can be applied for each type of scans to improve the quality of SVCT imaging and further assist in performance of downstream clinical usage. Extensive experiments across different scanning types provide validation for the effectiveness of task-specific sampling strategies in enhancing imaging quality. Experiments involving downstream tasks verify the clinical value of learned sampling strategies, as evidenced by notable improvements in downstream task performance. Furthermore, the utilization of a multi-task framework with a shared reconstruction network facilitates deployment on current imaging devices with switchable task-specific modules, and allows for easily integrate new tasks without retraining the entire model.

One of the main obstacles for the signal extraction of the three point correlation function using photometric surveys, such as the Rubin Observatory Legacy Survey of Space and Time (LSST), will be the prohibitive computation time required for dealing with a vast quantity of sources. Brute force algorithms, which naively scales as $\mathcal{O}(N^3)$ with the number of objects, can be further improved with tree methods but not enough to deal with large scale correlations of Rubin's data. However, a harmonic basis decomposition of these higher order statistics reduces the time dramatically, to scale as a two-point correlation function with the number of objects, so that the signal can be extracted in a reasonable amount of time. In this work, we aim to develop the framework to use these expansions within the Limber approximation for scalar (or spin-0) fields, such as galaxy counts, weak lensing convergence or aperture masses. We develop an estimator to extract the signal from catalogs and different phenomenological and theoretical models for its description. The latter includes halo model and standard perturbation theory, to which we add a simple effective field theory prescription based on the short range of non-locality of cosmic fields, significantly improving the agreement with simulated data. In parallel to the modeling of the signal, we develop a code that can efficiently calculate three points correlations of more than 200 million data points (a full sky simulation with Nside=4096) in $\sim$40 minutes on a single high-performance computing node, enabling a feasible analysis for the upcoming LSST data., Comment: 31 pages, 10 figures

We explain the large variety of star formation laws in terms of one single, simple law that can be inferred from the definition of the star formation rate and basic algebra. The resulting equation, $\SFR = \eff\ \Mcollapsing/\tauff$, although it has been presented elsewhere, is interpreted in terms of clouds undergoing collapse { rather than being turbulence-supported, an idea that different groups have pursued this century}. Under such assumption, one can explain the constancy of $\eff$, the different intra-cloud correlations observed in Milky Way's molecular clouds, as well as the resolved and unresolved extragalactic relationships between SFR and a measurement of the mass in CO, HCN, and CO+HI. We also explain why the slope of the correlation changes when the orbital time $\tauorb$ is considered instead of the free-fall time, and why estimations of the free-fall time from extragalactic observations skew the correlation, providing a false sublinear correlation. We furthermore show that the apparent nearly linear correlation between the star formation rate and the dynamical equilibrium pressure in the midplane of the galaxies, $\PDE$, is just a consequence of $\PDE$ values being dominated by the variation of the column density of molecular gas. All in all, we argue that the star formation law is driven by the collapse of cold, dense gas, which happens to be primarily molecular in the present Universe, and that the role of stellar feedback is just to shut down the star formation process, not to shape the star formation law., Comment: 17 pages, one figure. Accepted by MNRAS

Two subsequent deep learning networks, one localizing the fetal chest and one identifying a set of landmarks on a coronal whole-uterus balanced steady-state free precession scan, were trained on 167 and 71 fetal datasets across field strengths, acquisition parameters, and gestational ages and implemented in a real-time setup. Next, a phase-contrast sequence was modified to use the identified landmarks for planning. The OWL pipeline was evaluated retrospectively in 10 datasets and prospectively in 7 fetal subjects (gestational ages between 36+3 and 39+3 weeks). The prospective cases were additionally manually planned to enable direct comparison both qualitatively, by scoring the planning quality, and quantitatively, by comparing the indexed flow measurements. OWL enabled real-time fully automatic planning of the 2D phase-contrast scans in all but one of the prospective participants. The fetal body localization achieved an overall Dice score of 0.94+-0.05 and the cardiac landmark detection accuracy was 5.77+-2.91 mm for the descending aorta, 4.32+-2.44 mm for the spine, and 4.94+-3.82 mm for the umbilical vein. For the prospective cases, overall planning quality was 2.73/4 for the automated scans, compared to 3.0/4 for manual planning, and the flow quantitative evaluation showed a mean difference of -1.8% (range -14.2% to 14.9%) by comparing the indexed flow measurements obtained from gated automatic and manual acquisitions. Real-time automated planning of 2D phase-contrast MRI was effectively accomplished for 2 major vessels of the fetal vasculature. While demonstrated here on 0.55T, the achieved method has wider implications, and training across multiple field strengths enables generalization. OWL thereby presents an important step towards extending access to this modality beyond specialised centres., Comment: 12 pages

In the field of neuroimaging, accurate brain age prediction is pivotal for uncovering the complexities of brain aging and pinpointing early indicators of neurodegenerative conditions. Recent advancements in self-supervised learning, particularly in contrastive learning, have demonstrated greater robustness when dealing with complex datasets. However, current approaches often fall short in generalizing across non-uniformly distributed data, prevalent in medical imaging scenarios. To bridge this gap, we introduce a novel contrastive loss that adapts dynamically during the training process, focusing on the localized neighborhoods of samples. Moreover, we expand beyond traditional structural features by incorporating brain stiffness, a mechanical property previously underexplored yet promising due to its sensitivity to age-related changes. This work presents the first application of self-supervised learning to brain mechanical properties, using compiled stiffness maps from various clinical studies to predict brain age. Our approach, featuring dynamic localized loss, consistently outperforms existing state-of-the-art methods, demonstrating superior performance and laying the way for new directions in brain aging research.

We developed SaludConectaMX as a comprehensive system to track and understand the determinants of complications throughout chemotherapy treatment for children with cancer in Mexico. SaludConectaMX is unique in that it integrates patient clinical indicators with social determinants and caregiver mental health, forming a social-clinical perspective of the patient's evolving health trajectory. The system is composed of a web application (for hospital staff) and a mobile application (for family caregivers), providing the opportunity for cooperative patient monitoring in both hospital and home settings. This paper presents the system's preliminary design and usability evaluation results from a 1.5-year pilot study. Our findings indicate that while the hospital web app demonstrates high completion rates and user satisfaction, the family mobile app requires additional improvements for optimal accessibility; statistical and qualitative data analysis illuminate pathways for system improvement. Based on this evidence, we formalize suggestions for health system development in LMICs, which HCI researchers may leverage in future work.

Starting from the isomorphism between the AdS Carroll and Poincar\'e algebras, we map the three-dimensional asymptotically flat solutions of Poincar\'e gravity into an AdS Carroll spacetime. We show the mapped solutions satisfy the field equations of the Chern-Simons formulation of AdS Carroll gravity and exhibit a Carroll geometry structure. Despite the presence of a negative cosmological constant in the mapped spacetime, the algebra of the canonical generators of the asymptotic symmetries is given by the $\mathfrak{bms}_3$ algebra., Comment: 15 pages

We prove that a compact space $K$ embeds into a $\sigma$-product of compact metrizable spaces ($\sigma$-product of intervals) if and only if $K$ is (strongly countable-dimensional) hereditarily metalindel\"of and every subspace of $K$ has a nonempty relative open second-countable subset. This provides novel characterizations of $\omega$-Corson and $NY$ compact spaces. We give an example of a uniform Eberlein compact space that does not embed into a product of compact metric spaces in such a way that the $\sigma$-product is dense in the image. In particular, this answers a question of Kubi\'s and Leiderman. We also show that for a compact space $K$ the property of being $NY$ compact is determined by the topological structure of the space $C_p(K)$ of continuous real-valued functions of $K$ equipped with the pointwise convergence topology. This refines a recent result of Zakrzewski.

In this paper, we consider the problem of prototype-based vision-language reasoning problem. We observe that existing methods encounter three major challenges: 1) escalating resource demands and prolonging training times, 2) contending with excessive learnable parameters, and 3) fine-tuning based only on a single modality. These challenges will hinder their capability to adapt Vision-Language Models (VLMs) to downstream tasks. Motivated by this critical observation, we propose a novel method called NODE-Adapter, which utilizes Neural Ordinary Differential Equations for better vision-language reasoning. To fully leverage both visual and textual modalities and estimate class prototypes more effectively and accurately, we divide our method into two stages: cross-modal prototype construction and cross-modal prototype optimization using neural ordinary differential equations. Specifically, we exploit VLM to encode hand-crafted prompts into textual features and few-shot support images into visual features. Then, we estimate the textual prototype and visual prototype by averaging the textual features and visual features, respectively, and adaptively combine the textual prototype and visual prototype to construct the cross-modal prototype. To alleviate the prototype bias, we then model the prototype optimization process as an initial value problem with Neural ODEs to estimate the continuous gradient flow. Our extensive experimental results, which cover few-shot classification, domain generalization, and visual reasoning on human-object interaction, demonstrate that the proposed method significantly outperforms existing state-of-the-art approaches.

Regular screening and early discovery of uterine fibroid are crucial for preventing potential malignant transformations and ensuring timely, life-saving interventions. To this end, we collect and annotate the first ultrasound video dataset with 100 videos for uterine fibroid segmentation (UFUV). We also present Local-Global Reciprocal Network (LGRNet) to efficiently and effectively propagate the long-term temporal context which is crucial to help distinguish between uninformative noisy surrounding tissues and target lesion regions. Specifically, the Cyclic Neighborhood Propagation (CNP) is introduced to propagate the inter-frame local temporal context in a cyclic manner. Moreover, to aggregate global temporal context, we first condense each frame into a set of frame bottleneck queries and devise Hilbert Selective Scan (HilbertSS) to both efficiently path connect each frame and preserve the locality bias. A distribute layer is then utilized to disseminate back the global context for reciprocal refinement. Extensive experiments on UFUV and three public Video Polyp Segmentation (VPS) datasets demonstrate consistent improvements compared to state-of-the-art segmentation methods, indicating the effectiveness and versatility of LGRNet. Code, checkpoints, and dataset are available at https://github.com/bio-mlhui/LGRNet, Comment: MICCAI2024 Early Accept

Recent findings from DESI BAO combined with Planck CMB data have set an upper limit on the total neutrino mass of $\sum m_\nu < 0.072\, \text{eV}$ (95% confidence level), ruling out the inverted hierarchy. Indeed, methods that rely on the background expansion of the Universe tend to suggest negative neutrino masses. In this work, we contribute to the quest for accurately constraining neutrino mass using cosmological probes. By conducting a full-shape analysis on data from BOSS, eBOSS, and synthetic power spectra, we showed that projection effects can significantly influence constraints on neutrino mass, rendering these measurements largely unreliable. Our results highlight the need for better techniques to measure the neutrino mass accurately. Based on the large-scale structure suppression, we identified a critical blind spot in the full-shape analysis. By splitting the galaxy power spectrum into broadband and wiggles, we noticed that information on neutrino mass is primarily extracted from the suppressed wiggles rather than broadband suppression. This opens the possibility of developing alternative methods based only on the wiggles of the power spectrum that can be more robust than those heavily reliant on background evolution., Comment: 10 pages

Using methods of descriptive set theory, in particular, the determinacy of infinite games of perfect information, we answer several questions from the literature regarding different notions of bases in Banach spaces and lattices. For the case of Banach lattices, our results follow from a general theorem stating that (under the assumption of analytic determinacy), every $\sigma$-order basis $(e_n)$ for a Banach lattice $X=[e_n]$ is a uniform basis, and every uniform basis is Schauder. Moreover, the notions of order and $\sigma$-order bases coincide when $X=[e_n].$ Regarding Banach spaces, we address two problems concerning filter Schauder bases for Banach spaces, i.e., in which the norm convergence of partial sums is replaced by norm convergence along some appropriate filter on $\mathbb N$. We first provide an example of a Banach space admitting such a filter Schauder basis, but no ordinary Schauder basis. Secondly, we show that every filter Schauder basis with respect to an analytic filter is also a filter Schauder basis with respect to a Borel filter., Comment: A clarification on the coincidence of order basis and sigma-order basis for separable Banach lattices has been included

Image restoration is typically addressed through non-convex inverse problems, which are often solved using first-order block-wise splitting methods. In this paper, we consider a general type of non-convex optimisation model that captures many inverse image problems and present an inertial block proximal linearised minimisation (iBPLM) algorithm. Our new method unifies the Jacobi-type parallel and the Gauss-Seidel-type alternating update rules, and extends beyond these approaches. The inertial technique is also incorporated into each block-wise subproblem update, which can accelerate numerical convergence. Furthermore, we extend this framework with a plug-and-play variant (PnP-iBPLM) that integrates deep gradient denoisers, offering a flexible and robust solution for complex imaging tasks. We provide comprehensive theoretical analysis, demonstrating both subsequential and global convergence of the proposed algorithms. To validate our methods, we apply them to multi-block dictionary learning problems in image denoising and deblurring. Experimental results show that both iBPLM and PnP-iBPLM significantly enhance numerical performance and robustness in these applications., Comment: 6 figures, 6 tables

In this paper, part of the DREAMING Challenge - Diminished Reality for Emerging Applications in Medicine through Inpainting, we introduce a refined video inpainting technique optimised from the ProPainter method to meet the specialised demands of medical imaging, specifically in the context of oral and maxillofacial surgery. Our enhanced algorithm employs the zero-shot ProPainter, featuring optimized parameters and pre-processing, to adeptly manage the complex task of inpainting surgical video sequences, without requiring any training process. It aims to produce temporally coherent and detail-rich reconstructions of occluded regions, facilitating clearer views of operative fields. The efficacy of our approach is evaluated using comprehensive metrics, positioning it as a significant advancement in the application of diminished reality for medical purposes., Comment: Accepted to ISBI 2024

Deep learning has been extensively applied in medical image reconstruction, where Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) represent the predominant paradigms, each possessing distinct advantages and inherent limitations: CNNs exhibit linear complexity with local sensitivity, whereas ViTs demonstrate quadratic complexity with global sensitivity. The emerging Mamba has shown superiority in learning visual representation, which combines the advantages of linear scalability and global sensitivity. In this study, we introduce MambaMIR, an Arbitrary-Masked Mamba-based model with wavelet decomposition for joint medical image reconstruction and uncertainty estimation. A novel Arbitrary Scan Masking (ASM) mechanism "masks out" redundant information to introduce randomness for further uncertainty estimation. Compared to the commonly used Monte Carlo (MC) dropout, our proposed MC-ASM provides an uncertainty map without the need for hyperparameter tuning and mitigates the performance drop typically observed when applying dropout to low-level tasks. For further texture preservation and better perceptual quality, we employ the wavelet transformation into MambaMIR and explore its variant based on the Generative Adversarial Network, namely MambaMIR-GAN. Comprehensive experiments have been conducted for multiple representative medical image reconstruction tasks, demonstrating that the proposed MambaMIR and MambaMIR-GAN outperform other baseline and state-of-the-art methods in different reconstruction tasks, where MambaMIR achieves the best reconstruction fidelity and MambaMIR-GAN has the best perceptual quality. In addition, our MC-ASM provides uncertainty maps as an additional tool for clinicians, while mitigating the typical performance drop caused by the commonly used dropout.

The integration of predictive maintenance and cybersecurity represents a transformative advancement for small and medium-sized enterprises (SMEs) operating within the Industry 4.0 paradigm. Despite their economic importance, SMEs often face significant challenges in adopting advanced technologies due to resource constraints and knowledge gaps. The DETECTA 2.0 project addresses these hurdles by developing an innovative system that harmonizes real-time anomaly detection, sophisticated analytics, and predictive forecasting capabilities. The system employs a semi-supervised methodology, combining unsupervised anomaly detection with supervised learning techniques. This approach enables more agile and cost-effective development of AI detection systems, significantly reducing the time required for manual case review. At the core lies a Digital Twin interface, providing intuitive real-time visualizations of machine states and detected anomalies. Leveraging cutting-edge AI engines, the system intelligently categorizes anomalies based on observed patterns, differentiating between technical errors and potential cybersecurity incidents. This discernment is fortified by detailed analytics, including certainty levels that enhance alert reliability and minimize false positives. The predictive engine uses advanced time series algorithms like N-HiTS to forecast future machine utilization trends. This proactive approach optimizes maintenance planning, enhances cybersecurity measures, and minimizes unplanned downtimes despite variable production processes. With its modular architecture enabling seamless integration across industrial setups and low implementation costs, DETECTA 2.0 presents an attractive solution for SMEs to strengthen their predictive maintenance and cybersecurity strategies.

Partial differential equations (PDEs) with multiple scales or those defined over sufficiently large domains arise in various areas of science and engineering and often present problems when approximating the solutions numerically. Machine learning techniques are a relatively recent method for solving PDEs. Despite the increasing number of machine learning strategies developed to approximate PDEs, many remain focused on relatively small domains. When scaling the equations, a large domain is naturally obtained, especially when the solution exhibits multiscale characteristics. This study examines two-scale equations whose solution structures exhibit distinct characteristics: highly localized in some regions and significantly flat in others. These two regions must be adequately addressed over a large domain to approximate the solution more accurately. We focus on the vanishing gradient problem given by the diminishing gradient zone of the activation function over large domains and propose a stratified sampling algorithm to address this problem. We compare the uniform random classical sampling method over the entire domain and the proposed stratified sampling method. The numerical results confirm that the proposed method yields more accurate and consistent solutions than classical methods., Comment: 35 pages, 14 figures

Point cloud videos effectively capture real-world spatial geometries and temporal dynamics, which are essential for enabling intelligent agents to understand the dynamically changing 3D world we live in. Although static 3D point cloud processing has witnessed significant advancements, designing an effective 4D point cloud video backbone remains challenging, mainly due to the irregular and unordered distribution of points and temporal inconsistencies across frames. Moreover, recent state-of-the-art 4D backbones predominantly rely on transformer-based architectures, which commonly suffer from large computational costs due to their quadratic complexity, particularly when processing long video sequences. To address these challenges, we propose a novel 4D point cloud video understanding backbone based on the recently advanced State Space Models (SSMs). Specifically, our backbone begins by disentangling space and time in raw 4D sequences, and then establishing spatio-temporal correlations using our newly developed Intra-frame Spatial Mamba and Inter-frame Temporal Mamba blocks. The Intra-frame Spatial Mamba module is designed to encode locally similar or related geometric structures within a certain temporal searching stride, which can effectively capture short-term dynamics. Subsequently, these locally correlated tokens are delivered to the Inter-frame Temporal Mamba module, which globally integrates point features across the entire video with linear complexity, further establishing long-range motion dependencies. Experimental results on human action recognition and 4D semantic segmentation tasks demonstrate the superiority of our proposed method. Especially, for long video sequences, our proposed Mamba-based method has an 87.5% GPU memory reduction, 5.36 times speed-up, and much higher accuracy (up to +10.4%) compared with transformer-based counterparts on MSR-Action3D dataset.

In this study, we investigate the spin-charge current conversion characteristics of chemically disordered ferromagnetic single FePt thin films by spin-pumping ferromagnetic resonance experiments performed on both a resonance cavity and on patterned devices. We clearly observe a self-induced signal in a single FePt layer. The sign of a single FePt spin pumping voltage signal is consistent with a typical bilayer with a positive spin Hall angle layer such as that of Pt on top of a ferromagnet (FM), substrate//FM/Pt. Structural analysis shows a strong composition gradient due to natural oxidation at both FePt interfaces, with the Si substrate and with the air. The FePt-thickness dependence of the self-induced charge current produced allowed us to obtain $\lambda _ \text{FePt}=(1.5\pm 0.1)$ nm and self-induced $\theta_ \text{self-FePt}=0.047 \pm 0.003$, with efficiency for reciprocal effects applications $\theta _ \text{self-FePt} \times \lambda _ \text{FePt} = 0.071$ nm which is comparable to that of Pt, $\theta _ \text{SH-Pt} \times \lambda _ \text{Pt} = 0.2$ nm. Moreover, by studying bilayer systems such as Si//FePt/Pt and Si//Pt//FePt we independently could extract the individual contributions of the external inverse spin Hall effect of Pt and the self-induced inverse spin Hall effect of FePt. Notably, this method gives consistent values of charge currents produced due to only self-induced inverse spin Hall effect in FePt layers. These results advance our understanding of spin-to-charge interconversion mechanisms in composite thin films and pave the way for the development of next-generation spintronics devices based on self-torque., Comment: 13 pages, 7 figures

As network complexity escalates, there is an increasing need for more sophisticated methods to manage and operate these networks, focusing on enhancing efficiency, reliability, and security. A wide range of Artificial Intelligence (AI)/Machine Learning (ML) models are being developed in response. These models are pivotal in automating decision-making, conducting predictive analyses, managing networks proactively, enhancing security, and optimizing network performance. They are foundational in shaping the future of networks, collectively forming what is known as Network Intelligence (NI). Prominent Standard-Defining Organizations (SDOs) are integrating NI into future network architectures, particularly emphasizing the closed-loop approach. However, existing methods for seamlessly integrating NI into network architectures are not yet fully effective. This paper introduces an in-depth architectural design for a Network Intelligence Stratum (NI Stratum). This stratum is supported by a novel end-to-end NI orchestrator that supports closed-loop NI operations across various network domains. The primary goal of this design is to streamline the deployment and coordination of NI throughout the entire network infrastructure, tackling issues related to scalability, conflict resolution, and effective data management. We detail exhaustive workflows for managing the NI lifecycle and demonstrate a reference implementation of the NI Stratum, focusing on its compatibility and integration with current network systems and open-source platforms such as Kubernetes and Kubeflow, as well as on its validation on real-world environments. The paper also outlines major challenges and open issues in deploying and managing NI.

Understanding the large-scale three-dimensional structure of the inner heliosphere, while important in its own right, is crucial for space weather applications, such as forecasting the time of arrival and propagation of coronal mass ejections (CMEs). This study uses sunRunner3D (3D), a 3-D magnetohydrodynamic (MHD) model, to simulate solar wind (SW) streams and generate background states. SR3D employs the boundary conditions generated by CORona-HELiosphere (CORHEL) and the PLUTO code to compute the plasma properties of the SW with the MHD approximation up to 1.1 AU in the inner heliosphere. We demonstrate that SR3D reproduces global features of Corotating Interaction Regions (CIRs) observed by Earth-based spacecraft (OMNI) and the Solar TErrestial RElations Observatory (STEREO)-A for a set of Carrington rotations (CRs) that cover a period that lays in the late declining phase of solar cycle 24. Additionally, we demonstrate that the model solutions are valid in the corotating and inertial frames of references. Moreover, a comparison between SR3D simulations and in-situ measurements shows reasonable agreement with the observations, and our results are comparable to those achieved by Predictive Science Inc.'s Magnetohydrodynamic Algorithm outside a Sphere (MAS) code. We have also undertaken a comparative analysis with the Space Weather Adaptive Simulation Framework for Solar Wind (SWASTi-SW), a PLUTO physics-based model, to evaluate the precision of various initial boundary conditions. Finally, we discuss the disparities in the solutions derived from inertial and rotating frames., Comment: 31 pages, 9 figures, 3 tables, accepted for publication in the Journal of Space Weather and Space Climate

In anticipation of forthcoming data releases of current and future spectroscopic surveys, we present the validation tests and analysis of systematic effects within \texttt{velocileptors} modeling pipeline when fitting mock data from the \texttt{AbacusSummit} N-body simulations. We compare the constraints obtained from parameter compression methods to the direct fitting (Full-Modeling) approaches of modeling the galaxy power spectra, and show that the ShapeFit extension to the traditional template method is consistent with the Full-Modeling method within the standard $\Lambda$CDM parameter space. We show the dependence on scale cuts when fitting the different redshift bins using the ShapeFit and Full-Modeling methods. We test the ability to jointly fit data from multiple redshift bins as well as joint analysis of the pre-reconstruction power spectrum with the post-reconstruction BAO correlation function signal. We further demonstrate the behavior of the model when opening up the parameter space beyond $\Lambda$CDM and also when combining likelihoods with external datasets, namely the Planck CMB priors. Finally, we describe different parametrization options for the galaxy bias, counterterm, and stochastic parameters, and employ the halo model in order to physically motivate suitable priors that are necessary to ensure the stability of the perturbation theory., Comment: 56 pages, 23 figures. Supporting publication of DESI 2024 V: Analysis of the full shape of two-point clustering statistics from galaxies and quasars

DESI aims to provide one of the tightest constraints on cosmological parameters by analysing the clustering of more than thirty million galaxies. However, obtaining such constraints requires special care in validating the methodology and efforts to reduce the computational time required through data compression and emulation techniques. In this work, we perform a rigorous validation of the PyBird power spectrum modelling code with both a traditional emulated Full-Modelling approach and the model-independent ShapeFit compression approach. By using cubic box simulations that accurately reproduce the clustering and precision of the DESI survey, we find that the cosmological constraints from ShapeFit and Full-Modelling are consistent with each other at the $\sim0.5\sigma$ level for the $\Lambda$CDM model. Both ShapeFit and Full-Modelling are also consistent with the true $\Lambda$CDM simulation cosmology down to a scale of $k_{\mathrm{max}} = 0.20 h\mathrm{Mpc}^{-1}$ even after including the hexadecapole. For extended models such as the wCDM and the oCDM models, we find that including the hexadecapole can significantly improve the constraints and reduce the modelling errors with the same $k_{\mathrm{max}}$. While their discrepancies between the constraints from ShapeFit and Full-Modelling are more significant than $\Lambda$CDM, they remain consistent within $0.7\sigma$. Lastly, we also show that the constraints on cosmological parameters with the correlation function evaluated from PyBird down to $s_{\mathrm{min}} = 30 h^{-1} \mathrm{Mpc}$ are unbiased and consistent with the constraints from the power spectrum., Comment: Supporting publication of DESI 2024 V: Analysis of the full shape of two-point clustering statistics from galaxies and quasars (In prep). 51 pages, 21 figures, and 12 tables. 2nd revised version for JCAP

In preparation for the next generation of galaxy redshift surveys, and in particular the year-one data release from the Dark Energy Spectroscopic Instrument (DESI), we investigate the consistency of a variety of effective field theory models that describe the galaxy-galaxy power spectra in redshift space into the quasi-linear regime using 1-loop perturbation theory. These models are employed in the pipelines \texttt{velocileptors}, \texttt{PyBird}, and \texttt{Folps$\nu$}. While these models have been validated independently, a detailed comparison with consistent choices has not been attempted. After briefly discussing the theoretical differences between the models we describe how to provide a more apples-to-apples comparison between them. We present the results of fitting mock spectra from the \texttt{AbacusSummit} suite of N-body simulations provided in three redshift bins to mimic the types of dark time tracers targeted by the DESI survey. We show that the theories behave similarly and give consistent constraints in both the forward-modeling and ShapeFit compressed fitting approaches. We additionally generate (noiseless) synthetic data from each pipeline to be fit by the others, varying the scale cuts in order to show that the models agree within the range of scales for which we expect 1-loop perturbation theory to be applicable. This work lays the foundation of Full-Shape analysis with DESI Y1 galaxy samples where in the tests we performed, we found no systematic error associated with the modeling of the galaxy redshift space power spectrum for this volume., Comment: 33 pages, 12 figures. Supporting publication of DESI 2024 V: Analysis of the full shape of two-point clustering statistics from galaxies and quasars

The Dark Energy Spectroscopic Instrument (DESI) will provide unprecedented information about the large-scale structure of our Universe. In this work, we study the robustness of the theoretical modelling of the power spectrum of FOLPS, a novel effective field theory-based package for evaluating the redshift space power spectrum in the presence of massive neutrinos. We perform this validation by fitting the AbacusSummit high-accuracy $N$-body simulations for Luminous Red Galaxies, Emission Line Galaxies and Quasar tracers, calibrated to describe DESI observations. We quantify the potential systematic error budget of FOLPS, finding that the modelling errors are fully sub-dominant for the DESI statistical precision within the studied range of scales. Additionally, we study two complementary approaches to fit and analyse the power spectrum data, one based on direct Full-Modelling fits and the other on the ShapeFit compression variables, both resulting in very good agreement in precision and accuracy. In each of these approaches, we study a set of potential systematic errors induced by several assumptions, such as the choice of template cosmology, the effect of prior choice in the nuisance parameters of the model, or the range of scales used in the analysis. Furthermore, we show how opening up the parameter space beyond the vanilla $\Lambda$CDM model affects the DESI observables. These studies include the addition of massive neutrinos, spatial curvature, and dark energy equation of state. We also examine how relaxing the usual Cosmic Microwave Background and Big Bang Nucleosynthesis priors on the primordial spectral index and the baryonic matter abundance, respectively, impacts the inference on the rest of the parameters of interest. This paper pathways towards performing a robust and reliable analysis of the shape of the power spectrum of DESI galaxy and quasar clustering using FOLPS., Comment: Supporting publication of DESI 2024 VII: Cosmological constraints from full-shape analyses of the two-point clustering statistics measurements, in preparation (2024)

In the contemporary era of high-precision spectroscopic surveys, led by projects like DESI, there is an increasing demand for optimizing the extraction of cosmological information from clustering data. This work conducts a thorough comparison of various methodologies for modeling the full shape of the two-point statistics in configuration space. We investigate the performance of both direct fits (Full-Modeling) and the parameter compression approaches (ShapeFit and Standard). We utilize the ABACUS-SUMMIT simulations, tailored to exceed DESI's precision requirements. Particularly, we fit the two-point statistics of three distinct tracers (LRG, ELG, and QSO), by employing a Gaussian Streaming Model in tandem with Convolution Lagrangian Perturbation Theory and Effective Field Theory. We explore methodological setup variations, including the range of scales, the set of galaxy bias parameters, the inclusion of the hexadecapole, as well as model extensions encompassing varying $n_s$ and allowing for $w_0w_a$CDM dark energy model. Throughout these varied explorations, while precision levels fluctuate and certain configurations exhibit tighter parameter constraints, our pipeline consistently recovers the parameter values of the mocks within $1\sigma$ in all cases for a 1-year DESI volume. Additionally, we compare the performance of configuration space analysis with its Fourier space counterpart using three models: PyBird, FOLPS and velocileptors, presented in companion papers. We find good agreement with the results from all these models., Comment: Supporting publication of DESI 2024 KP5

We present cosmological results from the measurement of baryon acoustic oscillations (BAO) in galaxy, quasar and Lyman-$\alpha$ forest tracers from the first year of observations from the Dark Energy Spectroscopic Instrument (DESI), to be released in the DESI Data Release 1. DESI BAO provide robust measurements of the transverse comoving distance and Hubble rate, or their combination, relative to the sound horizon, in seven redshift bins from over 6 million extragalactic objects in the redshift range $0.1-1$ and $w_a<0$. This preference is 2.6$\sigma$ for the DESI+CMB combination, and persists or grows when SN~Ia are added in, giving results discrepant with the $\Lambda$CDM model at the $2.5\sigma$, $3.5\sigma$ or $3.9\sigma$ levels for the addition of Pantheon+, Union3, or DES-SN5YR datasets respectively. For the flat $\Lambda$CDM model with the sum of neutrino mass $\sum m_\nu$ free, combining the DESI and CMB data yields an upper limit $\sum m_\nu < 0.072$ $(0.113)$ eV at 95% confidence for a $\sum m_\nu>0$ $(\sum m_\nu>0.059)$ eV prior. These neutrino-mass constraints are substantially relaxed in models beyond $\Lambda$CDM. [Abridged.], Comment: This DESI Collaboration Key Publication is part of the 2024 publication series using the first year of observations (see https://data.desi.lbl.gov/doc/papers). Typos corrected and a new figure and discussion added to Appendix A

We present the measurement of Baryon Acoustic Oscillations (BAO) from the Lyman-$\alpha$ (Ly$\alpha$) forest of high-redshift quasars with the first-year dataset of the Dark Energy Spectroscopic Instrument (DESI). Our analysis uses over $420\,000$ Ly$\alpha$ forest spectra and their correlation with the spatial distribution of more than $700\,000$ quasars. An essential facet of this work is the development of a new analysis methodology on a blinded dataset. We conducted rigorous tests using synthetic data to ensure the reliability of our methodology and findings before unblinding. Additionally, we conducted multiple data splits to assess the consistency of the results and scrutinized various analysis approaches to confirm their robustness. For a given value of the sound horizon ($r_d$), we measure the expansion at $z_{\rm eff}=2.33$ with 2\% precision, $H(z_{\rm eff}) = (239.2 \pm 4.8) (147.09~{\rm Mpc} /r_d)$ km/s/Mpc. Similarly, we present a 2.4\% measurement of the transverse comoving distance to the same redshift, $D_M(z_{\rm eff}) = (5.84 \pm 0.14) (r_d/147.09~{\rm Mpc})$ Gpc. Together with other DESI BAO measurements at lower redshifts, these results are used in a companion paper to constrain cosmological parameters., Comment: This DESI Collaboration Key Publication is part of the 2024 publication series using the first year of observations (see https://data.desi.lbl.gov/doc/papers). Minor changes in v4, version accepted for publication in JCAP

We present the DESI 2024 galaxy and quasar baryon acoustic oscillations (BAO) measurements using over 5.7 million unique galaxy and quasar redshifts in the range 0.1

Assume $\text{MA}(\kappa)$. We show that for every real chain of size $\kappa$ in the quotient Boolean algebra $P(\omega)/fin$ we can find an almost chain of representatives such that every $n\in\omega$ oscillates at most three times along the almost chain. This is used to show that for every countable discrete extension of a separable compact line $K$ of weight $\kappa$ there exists an extension operator $E:C(K)\longrightarrow C(L)$ of norm at most three.

Segmentation of fetal brain tissue from magnetic resonance imaging (MRI) plays a crucial role in the study of in utero neurodevelopment. However, automated tools face substantial domain shift challenges as they must be robust to highly heterogeneous clinical data, often limited in numbers and lacking annotations. Indeed, high variability of the fetal brain morphology, MRI acquisition parameters, and superresolution reconstruction (SR) algorithms adversely affect the model's performance when evaluated out-of-domain. In this work, we introduce FetalSynthSeg, a domain randomization method to segment fetal brain MRI, inspired by SynthSeg. Our results show that models trained solely on synthetic data outperform models trained on real data in out-ofdomain settings, validated on a 120-subject cross-domain dataset. Furthermore, we extend our evaluation to 40 subjects acquired using lowfield (0.55T) MRI and reconstructed with novel SR models, showcasing robustness across different magnetic field strengths and SR algorithms. Leveraging a generative synthetic approach, we tackle the domain shift problem in fetal brain MRI and offer compelling prospects for applications in fields with limited and highly heterogeneous data., Comment: 10 pages, 5 figures, 1 table

Early detection of Alzheimer's disease's precursor stages is imperative for significantly enhancing patient outcomes and quality of life. This challenge is tackled through a semi-supervised multi-modal diagnosis framework. In particular, we introduce a new hypergraph framework that enables higher-order relations between multi-modal data, while utilising minimal labels. We first introduce a bilevel hypergraph optimisation framework that jointly learns a graph augmentation policy and a semi-supervised classifier. This dual learning strategy is hypothesised to enhance the robustness and generalisation capabilities of the model by fostering new pathways for information propagation. Secondly, we introduce a novel strategy for generating pseudo-labels more effectively via a gradient-driven flow. Our experimental results demonstrate the superior performance of our framework over current techniques in diagnosing Alzheimer's disease.

Recent advances in deep learning have significantly improved brain tumour segmentation techniques; however, the results still lack confidence and robustness as they solely consider image data without biophysical priors or pathological information. Integrating biophysics-informed regularisation is one effective way to change this situation, as it provides an prior regularisation for automated end-to-end learning. In this paper, we propose a novel approach that designs brain tumour growth Partial Differential Equation (PDE) models as a regularisation with deep learning, operational with any network model. Our method introduces tumour growth PDE models directly into the segmentation process, improving accuracy and robustness, especially in data-scarce scenarios. This system estimates tumour cell density using a periodic activation function. By effectively integrating this estimation with biophysical models, we achieve better capture of tumour characteristics. This approach not only aligns the segmentation closer to actual biological behaviour but also strengthens the model's performance under limited data conditions. We demonstrate the effectiveness of our framework through extensive experiments on the BraTS 2023 dataset, showcasing significant improvements in both precision and reliability of tumour segmentation., Comment: 11 pages, 4 figures and 1 table. Accepted by MICCAI2024

Real-world vision tasks frequently suffer from the appearance of unexpected adverse weather conditions, including rain, haze, snow, and raindrops. In the last decade, convolutional neural networks and vision transformers have yielded outstanding results in single-weather video removal. However, due to the absence of appropriate adaptation, most of them fail to generalize to other weather conditions. Although ViWS-Net is proposed to remove adverse weather conditions in videos with a single set of pre-trained weights, it is seriously blinded by seen weather at train-time and degenerates when coming to unseen weather during test-time. In this work, we introduce test-time adaptation into adverse weather removal in videos, and propose the first framework that integrates test-time adaptation into the iterative diffusion reverse process. Specifically, we devise a diffusion-based network with a novel temporal noise model to efficiently explore frame-correlated information in degraded video clips at training stage. During inference stage, we introduce a proxy task named Diffusion Tubelet Self-Calibration to learn the primer distribution of test video stream and optimize the model by approximating the temporal noise model for online adaptation. Experimental results, on benchmark datasets, demonstrate that our Test-Time Adaptation method with Diffusion-based network(Diff-TTA) outperforms state-of-the-art methods in terms of restoring videos degraded by seen weather conditions. Its generalizable capability is also validated with unseen weather conditions in both synthesized and real-world videos.