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.     

Feature selection methods for root-cause analysis among top-level product attributes

Manufacturing companies not only strive to deliver flawless products but also monitor product failures in the field to identify potential quality issues. When product failures occur, quality engineers must identify the root cause to improve any affected product and process. This root-cause analysis can be supported by feature selection methods that identify relevant product attributes, such as manufacturing dates with an increased number of product failures. In this paper, we present different methods for feature selection and evaluate their ability to identify relevant product attributes in a root-cause analysis. First, we compile a list of feature selection methods. Then, we summarize the properties of product attributes in warranty case data and discuss these properties regarding the challenges they pose for machine learning algorithms. Next, we simulate datasets of warranty cases, which emulate these product properties. Finally, we compare the feature selection methods based on these simulated datasets. In the end, the univariate filter information gain is determined to be a suitable method for a wide range of applications. The comparison based on simulated data provides a more general result than other publications, which only focus on a single use case. Due to the generic nature of the simulated datasets, the results can be applied to various root-cause analysis processes in different quality management applications and provide a guideline for readers who wish to explore machine learning methods for their analysis of quality data.     

The dynamic effect of Micro-MHD convection on bubble grown at a horizontal microelectrode

Relatively low efficiency is the biggest obstacle to the popularization of water electrolysis, which is a particularly feasible way to produce super-pure hydrogen. Imposing a magnetic field can increase the hydrogen production efficiency of water electrolysis. However, the enhancement's detailed mechanism still lacks an insightful understanding of the bubbles' micro vicinity. Our recent work aims to understand why the micro-magnetohydrodynamic (MHD) convection hinders single bubbles' detachment on the microelectrode. A water electrolysis experiment by microelectrode is performed under an electrode-normal magnetic field, and dynamic analysis of the single bubble growing on microelectrodes is performed. The variation of bubble diameter with time in the presence or absence of the magnetic field was measured, and the forces acting on the bubble were quantified. The result shows that the micro-MHD convection, induced by Lorentz force, can give rise to a downward hydrodynamic pressure force that will not appear in large-scale MHD convection. This force can be of the same magnitude as the surface tension, so it dramatically hinders bubbles' detachment. Besides, the Kelvin force provides a new potential way for further improving the efficiency of water electrolysis.     

Prediction of Chameleonic Efficiency

Chameleonic properties, i. e., the capacity of a molecule to hide polarity in non-polar environments and expose it in water, help achieving sufficient permeability and solubility for drug molecules with high MW. We present models of experimental measures of polarity for a set of 24 FDA approved drugs (MW 405-1113) and one PROTAC (MW 1034). Conformational ensembles in aqueous and non-polar environments were generated using molecular dynamics. A linear regression model that predicts chromatographic apparent polarity (EPSA) with a mean unsigned error of 10 Ų was derived based on separate terms for donor, acceptor, and total molecular SASA. A good correlation (R²=0.92) with an experimental measure of hydrogen bond donor potential, Δlog P_oct-tol, was found for the mean hydrogen bond donor SASA of the conformational ensemble scaled with Abraham's A hydrogen bond acidity. Two quantitative measures of chameleonic behaviour, the chameleonic efficiency indices, are introduced. We envision that the methods presented herein will be useful to triage designed molecules and prioritize those with the best chance of achieving acceptable permeability and solubility.     

Ablation behavior of thin-blade co-deposited C/Cx-SiCy composites under the influence of the complex fluid conditions of the oxyacetylene torch

This paper offers a new way of testing the ablation property of material under an oxyacetylene torch using a thin-blade specimen, which costs much less time to reach the maximum temperature and provides a harsh turbulence fluid field that's closer to reality. The thin-blade specimen experiences a higher turbulent intensity than the traditional disk-like specimen, leading to more efficient heat exchange. The fluid field simulation agrees with the testing results. In addition, we manage to synthesize the C/Cx-SiCy composites with the co-deposition chemical vapor infiltration (CVI) method. The C/Cx-SiCy composites exhibit a similar anti-ablation property as C/C composites and consist of enough SiC phase simultaneously, combining the advantages of both C/C composites and C/SiC composites. The thin-blade C/Cx-SiCy composites show a lower linear ablation rate (1.6 μm/s) than C/C composites (4.1 μm/s) and C/SiC composites (19.6 μm/s) during the oxyacetylene test. The glass layer formed on the surface of C/Cx-SiCy could cling to the bulk material instead of peeling off due to the high PyC content in the matrix could protect the SiO₂ from blowing away.     

Optimization of flow in additively manufactured porous columns with graded permeability

Chemical engineering systems often involve a functional porous medium, such as in catalyzed reactive flows, fluid purifiers, and chromatographic separations. Ideally, the flow rates throughout the porous medium are uniform, and all portions of the medium contribute efficiently to its function. The permeability is a property of a porous medium that depends on pore geometry and relates flow rate to pressure drop. Additive manufacturing techniques raise the possibilities that permeability can be arbitrarily specified in three dimensions, and that a broader range of permeabilities can be achieved than by traditional manufacturing methods. Using numerical optimization methods, we show that designs with spatially varying permeability can achieve greater flow uniformity than designs with uniform permeability. We consider geometries involving hemispherical regions that distribute flow, as in many glass chromatography columns. By several measures, significant improvements in flow uniformity can be obtained by modifying permeability only near the inlet and outlet.     

Mitochondrial and Autophagic Regulation of Adult Neurogenesis in the Healthy and Diseased Brain

Adult neurogenesis is a highly regulated process during which new neurons are generated from neural stem cells in two discrete regions of the adult brain: the subventricular zone of the lateral ventricle and the subgranular zone of the dentate gyrus in the hippocampus. Defects of adult hippocampal neurogenesis have been linked to cognitive decline and dysfunction during natural aging and in neurodegenerative diseases, as well as psychological stress-induced mood disorders. Understanding the mechanisms and pathways that regulate adult neurogenesis is crucial to improving preventative measures and therapies for these conditions. Accumulating evidence shows that mitochondria directly regulate various steps and phases of adult neurogenesis. This review summarizes recent findings on how mitochondrial metabolism, dynamics, and reactive oxygen species control several aspects of adult neural stem cell function and their differentiation to newborn neurons. It also discusses the importance of autophagy for adult neurogenesis, and how mitochondrial and autophagic dysfunction may contribute to cognitive defects and stress-induced mood disorders by compromising adult neurogenesis. Finally, I suggest possible ways to target mitochondrial function as a strategy for stem cell-based interventions and treatments for cognitive and mood disorders.     

Theoretical estimation of dielectric loss of oxide glasses using nonequilibrium molecular dynamics simulations

To theoretically explore amorphous materials with a sufficiently low dielectric loss, which are essential for next-generation communication devices, the applicability of a nonequilibrium molecular dynamics simulation employing an external alternating electric field was examined using alkaline silicate glass models. In this method, the dielectric loss is directly evaluated as the phase shift of the dipole moment from the applied electric field. This method enabled us to evaluate the dielectric loss in a wide frequency range from 1 GHz to 10 THz. It was observed that the dielectric loss reaches its maximum at a few THz. The simulation method was found to qualitatively reproduce the effects of alkaline content and alkaline type on the dielectric loss. Furthermore, it reasonably reproduced the effect of mixed alkalines on the dielectric loss, which was observed in our experiments on sodium and/or potassium silicate glasses. Alkaline mixing was thus found to reduce the dielectric loss.     

自适应融合层级特征的混合退化图像复原算法

多种退化类型混合的图像比单一类型的退化图像降质更严重,很难建立精确模型对其复原,研究端到端的神经网络算法是复原的关键.现有的基于操作选择注意力网络的算法(operation-wiseattentionnetwork,OWAN)虽然有一定的性能提升,但是其网络过于复杂,运行较慢,复原图像缺乏高频细节,整体效果也有提升的空间.针对这些问题,提出一种基于层级特征融合的自适应复原算法.该算法直接融合不同感受野分支的特征,增强复原图像的结构;用注意力机制对不同层级的特征进行动态融合,增加模型的自适应性,降低了模型冗余;另外,结合L1损失和感知损失,增强了复原图像的视觉感知效果.在DIV2K,BSD500等数据集上的实验结果表明,该算法无论是在峰值信噪比和结构相似性上的定量分析,还是在主观视觉质量方面,均优于OWAN算法,充分证明了该算法的有效性.