YOLSO: You Only Look Small Object

Small object detection is challenging and far from satisfactory. Most general object detectors suffer from two critical issues with small objects: (1) Feature extractor based on classification network cannot express the characteristics of small objects reasonably due to insufficient appearance information of targets and a large amount of background interference around them. (2) The detector requires a much higher location accuracy for small objects than for general objects. This paper proposes an effective and efficient small object detector YOLSO to address the above problems. For feature representation, we analyze the drawbacks in previous backbones and present a Half-Space Shortcut(HSSC) module to build a background-aware backbone. Furthermore, a coarse-to-fine Feature Pyramid Enhancement(FPE) module is introduced for layer-wise aggregation at a granular level to enhance the semantic discriminability. For loss function, we propose an exponential L1 loss to promote the convergence of regression, and a focal IOU loss to focus on prime samples with high classification confidence and high IOU. Both of them significantly improves the location accuracy of small objects. The proposed YOLSO sets state-of-the-art results on two typical small object datasets, MOCOD and VeDAI, at a speed of over 200 FPS. In the meantime, it also outperforms the baseline YOLOv3 by a wide margin on the common COCO dataset.     

AQP3 Increases Intercellular Cohesion in NSCLC A549 Cell Spheroids through Exploratory Cell Protrusions

Tumor cell aggregation is critical for cell survival following the loss of extracellular matrix attachment and dissemination. However, the underlying mechanotransduction of clustering solitary tumor cells is poorly understood, especially in non-small cell lung cancers (NSCLC). Here, we examined whether cell surface protrusions played an important role in facilitating the physical contact between floating cells detached from a substrate. We employed poly-2-hydroxyethyl methacrylate-based 3D culture methods to mimic in vivo tumor cell cluster formation. The suprastructural analysis of human NSCLC A549 cell spheroids showed that finger-like protrusions clung together via the actin cytoskeleton. Time-lapse holotomography demonstrated that the finger-like protrusions of free-floating cells in 3D culture displayed exploratory coalescence. Global gene expression analysis demonstrated that the genes in the organic hydroxyl transport were particularly enriched in the A549 cell spheroids. Particularly, the knockdown of the water channel aquaporin 3 gene (AQP3) impaired multicellular aggregate formation in 3D culture through the rearrangement of the actomyosin cytoskeleton. Moreover, the cells with reduced levels of AQP3 decreased their transmigration. Overall, these data indicate that cell detachment-upregulated AQP3 contributes to cell surface protrusions through actomyosin cytoskeleton remodeling, causing the aggressive aggregation of free-floating cells dependent on the property of the substratum and collective metastasis.     

Diphenyl-Methane Based Thyromimetic Inhibitors for Transthyretin Amyloidosis

Thyromimetics, whose physicochemical characteristics are analog to thyroid hormones (THs) and their derivatives, are promising candidates as novel therapeutics for neurodegenerative and metabolic pathologies. In particular, sobetirome (GC-1), one of the initial halogen-free thyromimetics, and newly synthesized IS25 and TG68, with optimized ADME-Tox profile, have recently attracted attention owing to their superior therapeutic benefits, selectivity, and enhanced permeability. Here, we further explored the functional capabilities of these thyromimetics to inhibit transthyretin (TTR) amyloidosis. TTR is a homotetrameric transporter protein for THs, yet it is also responsible for severe amyloid fibril formation, which is facilitated by tetramer dissociation into non-native monomers. By combining nuclear magnetic resonance (NMR) spectroscopy, computational simulation, and biochemical assays, we found that GC-1 and newly designed diphenyl-methane-based thyromimetics, namely IS25 and TG68, are TTR stabilizers and efficient suppressors of TTR aggregation. Based on these observations, we propose the novel potential of thyromimetics as a multi-functional therapeutic molecule for TTR-related pathologies, including neurodegenerative diseases.     

Some picture fuzzy Bonferroni mean operators with their application to multicriteria decision making

In this paper, we extend the Bonferroni mean (BM) operator with the picture fuzzy numbers (PFNs) to propose novel picture fuzzy aggregation operators and demonstrate their application to multicriteria decision making (MCDM). On the basis of the algebraic operational rules of PFNs and BM, we introduce some aggregation operators: the picture fuzzy Bonferroni mean, the picture fuzzy normalized weighted Bonferroni mean, and the picture fuzzy ordered weighted Bonferroni mean. Then, a new picture fuzzy MCDM method is proposed with the help of the proposed operators. Lastly, a practical application of proposed model is given to verify the developed model and related results of the proposed model is compared with the results of the existing models to indicate its applicability.     

Decomposition of first-order hybrid Petri nets for hierarchical control of manufacturing systems

Although hybrid Petri net (HPN) is a popular formalism in modelling hybrid production systems, the HPN model of large scale systems gets substantially complicated for analysis and control due to large dimensionality of such systems. To overcome this problem, a typical approach is to decompose the net into subnets and then control the plant through hierarchical or decentralized structures. Although this concept has been widely discussed in the literature for discrete PNs, there is a lack of research for HPNs. In this paper, a new method of decomposition of first-order hybrid Petri nets (FOHPNs) is proposed first and then the hierarchical control of the subnets through a coordinator is introduced. The advantage of using the proposed approach is validated by an existing example. A sugar milling case study is analysed by using a decomposed FOHPN model and the optimization results are compared against the results of the approaches presented in other papers. Simulation results show not only an improvement in production rate, but also show the ability to control the plant online. In addition, by using the hierarchical control structure for an FOHPN model, it is possible to reduce the cost of communication links, improve the reliability of the system, maintain the plant locally, and partially redesign the system.     

Hierarchical multi-reservoir optimization modeling for real-world complexity with application to the Three Gorges system

High dimensionality in real-world multi-reservoir systems greatly hinders the application and popularity of evolutionary algorithms, especially for systems with heterogeneous units. An efficient hierarchical optimization framework is presented for search space reduction, determining the best water distributions, not only between cascade reservoirs, but also among different types of hydropower units. The framework is applied to the Three Gorges Project (TGP) system and the results demonstrate that the difficulties of multi-reservoir optimization caused by high dimensionality can be effectively solved by the proposed hierarchical method. For the day studied, power output could be increased by 6.79 GWh using an optimal decision with the same amount of water actually used; while the same amount of power could be generated with 2.59 × 10⁷ m³ less water compared to the historical policy. The methodology proposed is general in that it can be used for other reservoir systems and other types of heterogeneous unit generators.     

Wet chemical synthesis of Gd+3 doped vanadium Oxide/MXene based mesoporous hierarchical architectures as advanced supercapacitor material

In this paper, Gd³⁺ doped V₂O₅/Ti₃C₂T_x MXene (GVO/MX) hierarchical architectures have been synthesized by wet chemical approach. As prepared GVO/MX composite, along undoped VO and unsupported GVO were well characterized by XRD, FESEM, EDX, FT-IR and BET techniques. Electrochemical performance of VO, GVO and GVO/MX was evaluated by CV, GCD and EIS measurements. Among the three electrodes, GVO/MX composite exhibited highest electrochemical activity with the optimum specific capacitance of 1024 Fg^-1 at 10 mVs^?1. The specific capacitance of GVO/MX was ～1.7 and ～3 times higher than unsupported GVO (585 Fg^-1) and VO (326 Fg^-1), respectively. The cyclic life of GVO/MX with capacitance retention 96.12% was observed at 60 mVs^?1. EIS measurements showed reduction in electrochemical impedance for GVO/MX as compared to GVO and VO. The corresponding impedance values of Rct and Resr for GVO/MX were calculated as 18 Ω and 1.8 Ω, respectively. The superior capacitive ability of GVO/MX can be ascribed to its unique morphology, short diffusion path and high surface area of fabricated composite. Considering it, the present work provides a feasible strategy to fabricate highly effective electrode materials for next generation energy storage devices.     

Novel core-shell-like Ni-supported hierarchical beta zeolite catalysts on bioethanol steam reforming

Hierarchical-Beta zeolites have been hydrothermally synthesized by adding a new gemini organic surfactant. The used gemini surfactant play the role of a “pore-forming agents” on the mesoscale, on the same time, providing alkaline environment for the system. With this hierarchical Beta zeolite as the core support, we successfully prepared a shell layer of Ni-containing (22 wt%) petal-like core-shell-like catalyst and applied it to bioethanol steam reforming. At the reaction temperature of 350 °C–550 °C, the conversion rate of ethanol and the selectivity of hydrogen were always above 85% and 70%. After reaction of 100 h on stream at 400 °C, there were not obvious inactivation could be observed on NiNPs/OH-MBeta catalyst.     

Morphological evolution on the surface of hydrothermally synthesized hydroxyapatite microspheres in the presence of EDTMP

Well-ordered and surface engineered hierarchical hydroxyapatite microspheres (HAM) were prepared via a template free hydrothermal process. Ethylene diamine tetra (methylene phosphonic acid) (EDTMP) was used as chelating or regulating agent for the first time in this study. The results indicated the formation of sheet-like particles in the absence of EDTMP. On the other hand, microspheres with radially grown nanorods (HAMNR) or nanosheets (HAMNS) on the surface were obtained (with average diameter of 5?µm) in the presence of EDTMP. X-ray diffraction (XRD) and Fourier transform infrared (FTIR) spectroscopy were used to characterize the crystalline phases in the synthesized samples. Scanning electron microscopy (SEM) and transmission electron microscopy (TEM) showed that EDTMP concentration played an important part in regulating the morphology to form well organized microspheres with nanosheets or nanorods on the surface. Brunauer-Emmett-Teller (BET) revealed an increase in the specific surface area with the change in morphology from the HAMNS to HAMNR. Possible mechanisms are proposed to account for the formation of different morphologies based upon thermodynamic and kinetic theories.     

Model Selection for Unsupervised Learning of Visual Context

This study addresses the problem of choosing the most suitable probabilistic model selection criterion for unsupervised learning of visual context of a dynamic scene using mixture models. A rectified Bayesian Information Criterion (BICr) and a Completed Likelihood Akaike’s Information Criterion (CL-AIC) are formulated to estimate the optimal model order (complexity) for a given visual scene. Both criteria are designed to overcome poor model selection by existing popular criteria when the data sample size varies from small to large and the true mixture distribution kernel functions differ from the assumed ones. Extensive experiments on learning visual context for dynamic scene modelling are carried out to demonstrate the effectiveness of BICr and CL-AIC, compared to that of existing popular model selection criteria including BIC, AIC and Integrated Completed Likelihood (ICL). Our study suggests that for learning visual context using a mixture model, BICr is the most appropriate criterion given sparse data, while CL-AIC should be chosen given moderate or large data sample sizes.