All-in-one image restoration aims to recover clean images from observations corrupted by diverse and unknown degradations within a unified modeling framework. Existing approaches generally adopt a shared model for different degradation types, while neglecting the intrinsic differences in degradation complexity, thereby leading to inefficient computation and suboptimal restoration performance. In this paper, we revisit the image restoration problem from a structural perspective and reveal that complex degradations can be decomposed into simpler components and additional residual terms, forming an inherent hierarchical structure. This observation provides a principled understanding of task-wise complexity and motivates the design of adaptive restoration models. Based on this formulation, we propose a Unified-Width Adaptive Dynamic Network (U-WADN), which dynamically allocates computational resources according to the underlying degradation complexity. Specifically, a Width-Adaptive Backbone (WAB) is constructed to provide a hierarchy of subnetworks with different representational capacities, while a Width Selector (WS) is introduced to adaptively determine the appropriate subnetwork for each input. Extensive experiments across a variety of image restoration tasks demonstrate that the proposed U-WADN achieves better performance while simultaneously reducing up to 32.3% of FLOPs and providing approximately 15.7% real-time acceleration. The code has been made available at https://github.com/xuyimin0926/U-WADN.
Let g be the aiffine Lie superalgebra associated with g = sl(2|1)(C). In this paper, we obtain a character formula for all irreducible l-highest weight (g) over tilde -modules with fifinite-dimensional weight spaces, together with an explicit construction of these modules. To this end, we establish a necessary and sufficient criterion for the irreducibility of all parabolically induced modules over the corresponding r-th Takiffff superalgebra of g. (c) 2026 Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
Postoperative peritendinous adhesion arises from sustained fibroblast activity, collagen overproduction, and dysregulated mechanical signaling during tendon repair; however, few current treatments can deliver a prolonged molecular intervention across the proliferative stage of healing. Here, we developed a fluorinated covalent organic framework hydrogel, Thi@FCOF@Gel, that enables temporally controlled delivery of thioridazine hydrochloride for sustained prevention of peritendinous adhesion. Through phenotypic drug screening, thioridazine hydrochloride (Thi) is identified as a previously unrecognized anti-adhesion agent capable of selectively suppressing fibroblast proliferation and myofibroblast activation while exerting limited inhibitory effects on tendon-derived stem cells. To synchronize drug exposure with adhesion formation, thioridazine was encapsulated into a fluorinated COF and incorporated into a hyaluronic acid/quaternized chitosan hydrogel, yielding an injectable, self-healing and mechanically reinforced local depot with prolonged release behavior. Mechanistically, thioridazine-containing hydrogels attenuated extracellular matrix-associated mechanosignaling by downregulating Piezo1, increasing YAP phosphorylation and suppressing α-SMA and collagen III expression. In a rat Achilles tendon repair model, Thi@FCOF@Gel markedly reduced fibrotic bridging, collagen deposition and adhesion scores, while improving gait recovery without compromising tendon biomechanical integrity or systemic safety. This study not only reveals a new anti-adhesion application of Thi, but also provides a COF-enabled hydrogel strategy for precise local drug delivery in postoperative adhesion management.
To address the lack of quantitative reliability criteria for the architecture selection of aviation hybrid electric propulsion system (HEPS), this paper conducts a reliability-mass trade-off analysis for a 200 kW-class series HEPS, considering its multi-flight-state characteristics and stringent lightweight requirements. Three typical architectures are comprehensively evaluated. The main findings are as follows: First, the per-flight-hour failure probabilities of the centralized, redundant, and decentralized architectures are 3.9 & times; 10-9, 1.7 & times; 10-11, and 1.2 & times; 10-11, respectively. The centralized architecture fails to meet the catastrophic safety objective (10-9) specified in EASA SC-VTOL-01. Second, under the time-based weighting evaluation, the decentralized architecture exhibits the highest relative reliability upon reaching a dynamic equilibrium. Nevertheless, after introducing a mass penalty, the unit reliability of the redundant architecture is about 2.2% higher than that of the decentralized one, showing an optimal trade-off between reliability and mass. Overall, this study provides a reliability evaluation framework incorporating a mass penalty, offering technical support and reference solutions for the design of the aviation HEPS with high specific power and high reliability.
