Human action understanding aims to detect action instances and classify them in videos. Existing methods capture contextual dependencies through temporal feature aggregation or attention-based interactions. However, these methods primarily rely on visual features to establish relationships, making them susceptible to interference from background content or irrelevant action frames. To address this limitation, we propose Hybrid Causal Action Reasoning (HCAR), a framework for modeling causal correlations between action proposals. Specifically, we design a Hybrid Causality-Aware Attention (HCAA) to refine the semantic representations of proposal classes. To further characterize relations between adjacent proposals, an ordered relation branch evaluates the refined representations in temporal order, while a hyperbolic branch measures hierarchical semantic relatedness between action categories in the Poincaré ball. The outputs of these branches are combined to obtain causal correlation scores, which are subsequently used to refine action proposals. Extensive experiments on THUMOS14 and ActivityNet-1.3 demonstrate the effectiveness of our HCAR.
Collagen is one of the main constituents of cosmetic formulations because of its moisturizing, regenerating, and film-forming properties. The aim of the present study was to obtain biocompatible collagen hydrolysates from Hypophthalmichthys molitrix skin, using a neutral inexpensive hydrolysis process involving low temperature (135 °C) and pressure values (315 kPa) and by using various pretreatment procedures (in order to eliminate the scales, fat, and meat). Comprehensive characterization of the hydrolysate samples confirmed their high purity, evidenced by a protein content exceeding 10%, undetected ash levels, and a pH between 6 and 7. This was further supported by optical microscopy and SEM imaging, which revealed the characteristic fibril structure. Additionally, FTIR analysis and CD spectroscopy verified a specific random coil structure and the complete absence of the triple helix characteristic of denatured collagen. EDX results confirmed the presence of C, O, N, and traces of S in all collagen hydrolysate samples. The results of the microbiological analysis for E. coli and S. aureus show either the complete absence or negligible amounts of fungi and bacteria in all hydrolysate samples, thereby demonstrating the effectiveness of the extraction process. The in vitro biocompatibility assessment established a non-cytotoxic concentration range between 1.25 mg/mL and 10 mg/mL for the tested samples, with cell viability above 70% and LDH release below 5%. Furthermore, clinical patch tests of the obtained fish collagen hydrolysates revealed no irritation reactions. Neutral hydrolysis of silver carp skin collagen hydrolysates with determined characteristics and favorable biocompatibility proves to be suitable for cosmetic applications.
Axonal growth is guided by biochemical, mechanical, and geometrical cues while remaining subject to stochastic fluctuations associated with cytoskeletal dynamics, adhesion, signaling, and growth cone sensing. Here, we develop two complementary variational descriptions of mechanically guided axonal growth. First, for neurons growing on periodic parallel micropatterns, we use previously established Langevin and Fokker–Planck descriptions to derive a stochastic path action that combines deterministic substrate guidance with random fluctuations. The probability of a growth trajectory decreases exponentially with this action, so that, for fixed boundary conditions, the most probable axonal trajectories correspond to minimum action paths. Second, for axons growing on substrates with spatially varying stiffness, we show that stiffness-dependent traction can generate a mechanical guidance response. At an interface between regions of different stiffness, this traction-dependent mechanical response produces axonal turning and provides the basis for a variational description of refraction. In a quasi-static limit, the same mechanical framework leads to a Fermat-like spatial action with an effective axonal refractive index determined by the traction–stiffness relationship. This formulation predicts refraction of axons across stiffness boundaries and yields an approximate Snell-like relation and critical angle condition. The stochastic and Fermat-like actions have different physical meanings and ranges of validity, but both provide path-based descriptions of how substrate-dependent mechanical cues constrain axonal trajectories. These results connect measurable quantities including drift, diffusion, substrate stiffness, traction, and growth direction to global properties of axonal paths and suggest new strategies for designing biomimetic substrates for controlled neuronal growth and regeneration.
Flexible biological features, such as bird covert feathers and fish fins, provide remarkable flow control without external energy input. Inspired by nature, flow control using flexible auxiliary structures has attracted substantial scientific interest in recent years. They modify flow fields through morphological reconfiguration and flow-induced flapping, thereby modulating energy exchange between the fluid and the structure. Recent research can be categorized into five major applications: drag reduction, vibration suppression, lift augmentation, noise mitigation, and mixing enhancement. This review synthesizes and critically evaluates existing studies on flexible auxiliary structures across these five application areas, with particular emphasis on the underlying flow-control mechanisms. It further analyses how key parameters, including bending rigidity, geometric dimensions, and installation position, influence the control performance. It also distills the primary challenges facing flexible auxiliary structures as they transition from theoretical research to practical applications and outlines prospects for their evolution towards intelligent control systems.
Industrial anomaly detection is a critical component of machine-vision-based intelligent manufacturing, especially when defective samples are scarce and pixel-level annotations are expensive to obtain. PaDiM is a widely used unsupervised anomaly detection method that models patch-level feature distributions extracted from pretrained convolutional neural networks; its practical performance depends on the selected feature layers and the dimensionality of patch embeddings. This study proposes FO-PaDiM, a feature configuration optimization framework built on the PaDiM pipeline that keeps the original statistical modeling unchanged while exhaustively searching feature-layer combinations and embedding dimensionality. Unlike a previous exploratory version of this study, all configuration selection is performed under a leakage-free development protocol—a deterministic 80/20 split of training normals combined with multi-type synthetic anomalies—so that the MVTec AD test set is used exactly once, with a locked configuration and frozen thresholds. On eight MVTec AD categories (bottle, cable, capsule, hazelnut, metal_nut, pill, screw, transistor) with 14 configurations per category (seven non-empty ResNet-18 layer combinations × n_features in {32, 64}), the selection is found to be category-dependent: single-stage configurations (layer2, layer3 or layer1) are preferred in six categories and multi-stage combinations in two, while the previously reported universal preference for the three-layer combination does not replicate once test-set leakage is removed. Under the four evaluated weighting schemes, the layer combination remains unchanged for 62.5% of categories and n_features for 87.5%. Final locked test performance reaches a mean Image AUROC of 0.8676, Image F1 of 0.8079, Pixel AUROC of 0.9644, and Pixel F1 of 0.4582. The results provide evidence that multi-level feature configurations can be beneficial but that their advantage is category-dependent and small in magnitude, and they confirm the persistent difficulty of threshold-dependent binary localization (Pixel F1) despite strong ranking quality (Pixel AUROC).
