A Survey on Long-Tailed Visual Recognition

The heavy reliance on data is one of the major reasons that currently limit the development of deep learning. Data quality directly dominates the effect of deep learning models, and the long-tailed distribution is one of the factors affecting data quality. The long-tailed phenomenon is prevalent due to the prevalence of power law in nature. In this case, the performance of deep learning models is often dominated by the head classes while the learning of the tail classes is severely underdeveloped. In order to learn adequately for all classes, many researchers have studied and preliminarily addressed the long-tailed problem. In this survey, we focus on the problems caused by long-tailed data distribution, sort out the representative long-tailed visual recognition datasets and summarize some mainstream long-tailed studies. Specifically, we summarize these studies into ten categories from the perspective of representation learning, and outline the highlights and limitations of each category. Besides, we have studied four quantitative metrics for evaluating the imbalance, and suggest using the Gini coefficient to evaluate the long-tailedness of a dataset. Based on the Gini coefficient, we quantitatively study 20 widely-used and large-scale visual datasets proposed in the last decade, and find that the long-tailed phenomenon is widespread and has not been fully studied. Finally, we provide several future directions for the development of long-tailed learning to provide more ideas for readers.

     

Occluded Video Instance Segmentation: A Benchmark

International Journal of Computer Vision - Can our video understanding systems perceive objects when a heavy occlusion exists in a scene? To answer this question, we collect a large-scale dataset...     

世界主要国家工程科技重大计划与前沿问题综述

本文梳理了近年来世界各主要国家工程科技领域的重大战略计划和措施,简要分析了当前工程科技主要领域发展状况、水平,对主要国家工程科技领域计划的目标、路径与内容进行了比较分析,描绘了工程科技计划关系图、聚焦点和前沿问题。     

Ultrathin Co3O4 nanosheets highly dispersed on reduced graphene oxide: An efficient bi-functional catalyst for Li-O2 battery

Ultrathin Co₃O₄ nanosheets grown on the reduced graphene oxide (Co₃O₄/rGO) was synthesized by a simple hydrothermal method and was investigated as a cathode in a Li-O₂ battery. Benefited from the synergistic effect between Co₃O₄ and rGO, the hybrid exhibits a high initial capacity of 10,528 mAh g^?1 along with a high coulombic efficiency (84.4%) at 100 mA g^?1. In addition, the batteries show an enhanced cycling stability and after 113 cycles, the cut-off discharge voltage remains above 2.5 V. The outstanding performance is intimately related to the high surface area of rGO, which not only provide carbon skeleton for the uniform distribution of Co₃O₄ nanosheets but also facilitate the reversible formation and decomposition of insoluble Li₂O₂. The results of electrochemical tests confirm that the Co₃O₄/rGO hybrid is a promising candidate for the Li-O₂ batteries.     

Structure and Magnetic Properties of Ce-Substituted Yttrium Iron Garnet Prepared by Conventional Sintering Techniques

Y_3?xCe _x Fe₅ O ₁₂ (CeYIG) ceramics, with x = 0, 0.15, 0.25, 0.35, 0.45, and 0.5, were fabricated by a conventional ceramic sintering technique. We studied the structures and magnetic fields of a series of CeYIG ceramics using X-ray powder diffraction, a scanning electron microscope, and a superconducting quantum interference device magnetometer. Findings showed that the substitution limit of the concentration of Ce³⁺ ions in the yttrium iron garnet structure was approximately x = 0.25. An extra CeO₂ phase was detected in the ceramic when the addition of CeO₂ content overtook the limit. The lattice constants and relative densities increased by increasing the Ce³⁺ contents in the ceramics. First, the saturation magnetization increased gradually with increases in the substitute concentration of Ce³⁺ ions and then decreased gradually when x = 0.35, 0.45, and 0.5. Overall, this study showed that the Y_3?xCe _x Fe₅ O ₁₂ material with x ≤ 0.15 exhibited excellent magnetic properties. Hence, the material show promise for magneto-optical and microwave communication applications.     

Ferromagnetism and the Optical Properties of Mn-Doped CdSe with the Wurtzite Structure

The geometrical structure of CdSe was optimized by using the ultrasoft pseudopotential method of a total energy plane wave based on density functional theory. The band structure, density of states, and optical properties were calculated and discussed in detail. The Mn-doped CdSe is found to be a half-metallic ferromagnet with 100% carrier spin polarization at the Fermi level. At a Mn concentration of 12.5%, the calculated total energy of the spin-polarized state is 614 meV lower than that of the nonspin-polarized state. The net magnetic moment of 5 μ _B is found per supercell for 12.5% Mn-doped CdSe. The estimated Curie temperature of 748.6 K for Mn-doped CdSe is above room temperature. The ferromagnetic ground state in Mn-doped CdSe can be explained in terms of the p ? d hybridization mechanism. These results suggest that Mn-doped CdSe may present a promising dilute magnetic semiconductor, and may have potential applications in the field of spintronics.     

Drug‐Loaded Mesoporous Tantalum Oxide Nanoparticles for Enhanced Synergetic Chemoradiotherapy with Reduced Systemic Toxicity

Combining chemotherapy and radiotherapy (chemoradiotherapy) has been widely applied in many clinical practices, showing promises in enhancing therapeutic outcomes. Nontoxic nanocarriers that not only are able to deliver chemotherapeutics into tumors, but could also act as radiosensitizers to enhance radiotherapy would thus be of great interest in the development of chemoradiotherapies. To achieve this aim, herein mesoporous tantalum oxide (mTa₂O₅) nanoparticles with polyethylene glycol (PEG) modification are fabricated. Those mTa₂O₅‐PEG nanoparticles could serve as a drug delivery vehicle to allow efficient loading of chemotherapeutics such as doxorubicin (DOX), whose release appears to be pH responsive. Meanwhile, owing to the interaction of Ta with X‐ray, mTa₂O₅‐PEG nanoparticles could offer an intrinsic radiosensitization effect to increase X‐ray‐induced DNA damages during radiotherapy. As a result, DOX‐loaded mTa₂O₅‐PEG (mTa₂O₅‐PEG/DOX) nanoparticles can offer a strong synergistic therapeutic effect during the combined chemoradiotherapy. Furthermore, in chemoradiotherapy, such mTa₂O₅‐PEG/DOX shows remarkably reduced side effects compared to free DOX, which at the same dose appears to be lethal to animals. This work thus presents a new type of mesoporous nanocarrier particularly useful for the delivery of safe and effective chemoradiotherapy.     

Structure and Magnetic Property Control of Copper Hydroxide Acetate by Non‐Classical Crystallization

Copper hydroxide acetate (CHA), one layered hydroxide compound with tunable magnetism, attracts great interest because of its potential applications in memory devices. However, ferromagnetism for CHA is only demonstrated by means of GPa pressure. Herein, a new method is reported, involving the combination of different crystallization pathways to control crystallization of amorphous CHA toward the formation of CHA/polymer composites with tunable magnetic properties and even a tunability that can be tested at room temperature. By using poly[(ethylene glycol)₆ methyl ether methacrylate]‐block‐poly[2‐(acetoacetoxy) ethyl methacrylate] (PEGMA‐b‐PAEMA) diblock copolymers as additives in combination with a post‐treatment process by ultracentrifugation, it is demonstrated that CHA and PEGMA‐b‐PAEMA form composites exhibiting different magnetic properties, depending on CHA in‐plane nanostructures. Analytical characterization reveals that crystallization of CHA is induced by ultracentrifugation, during which CHA nanostructures can be well controlled by changing the degrees of polymerization of the PEGMA and PAEMA blocks and their block length ratios. These findings not only present the first example of using crystallization from polymer stabilized amorphous precursors toward the generation of magnetic nanomaterials with tunable magnetism but also pave the way for the future design of functional composite materials.     

Mesoporous Germanium Anode Materials for Lithium‐Ion Battery with Exceptional Cycling Stability in Wide Temperature Range

Porous structured materials have unique architectures and are promising for lithium‐ion batteries to enhance performances. In particular, mesoporous materials have many advantages including a high surface area and large void spaces which can increase reactivity and accessibility of lithium ions. This study reports a synthesis of newly developed mesoporous germanium (Ge) particles prepared by a zincothermic reduction at a mild temperature for high performance lithium‐ion batteries which can operate in a wide temperature range. The optimized Ge battery anodes with the mesoporous structure exhibit outstanding electrochemical properties in a wide temperature ranging from ?20 to 60 °C. Ge anodes exhibit a stable cycling retention at various temperatures (capacity retention of 99% after 100 cycles at 25 °C, 84% after 300 cycles at 60 °C, and 50% after 50 cycles at ?20 °C). Furthermore, full cells consisting of the mesoporous Ge anode and an LiFePO₄ cathode show an excellent cyclability at ?20 and 25 °C. Mesoporous Ge materials synthesized by the zincothermic reduction can be potentially applied as high performance anode materials for practical lithium‐ion batteries.