Representing features at multiple scales is of great importance for numerous vision tasks. Recent advances in backbone convolutional neural networks (CNNs) continually demonstrate stronger multi-scale representation ability, leading to consistent performance gains on a wide range of applications. However, most existing methods represent the multi-scale features in a layer-wise manner. In this paper, we propose a novel building block for CNNs, namely Res2Net, by constructing hierarchical residual-like connections within one single residual block. The Res2Net represents multi-scale features at a granular level and increases the range of receptive fields for each network layer. The proposed Res2Net block can be plugged into the state-of-the-art backbone CNN models, e.g., ResNet, ResNeXt, and DLA. We evaluate the Res2Net block on all these models and demonstrate consistent performance gains over baseline models on widely-used datasets, e.g., CIFAR-100 and ImageNet. Further ablation studies and experimental results on representative computer vision tasks, i.e., object detection, class activation mapping, and salient object detection, further verify the superiority of the Res2Net over the state-of-the-art baseline methods. The source code and trained models are available on https://mmcheng.net/res2net/.

ZnO, aside from TiO2, has been considered as a promising material for purification and disinfection of water and air, and remediation of hazardous waste, owing to its high activity, environment-friendly feature and lower cost. However, their poor visible light utilization greatly limited their practical applications. Herein, we demonstrate the fabrication of different aspect ratios of the ZnO nanorods with surface defects by mechanical-assisted thermal decomposition method. The experiments revealed that ZnO nanorods with higher aspect ratio and surface defects show significantly higher photocatalytic performances.

Seeding a conventional chemical oxidative polymerization of aniline with even very small amounts of biological, inorganic, or organic nanofibers (usually <1%) dramatically changes the morphology of the resulting doped electronic polymer polyaniline from nonfibrillar (particulate) to almost exclusively nanofibers. The nanoscale morphology of the original seed template is transcribed almost quantitatively to the bulk precipitate. These findings could have immediate impact in the design and development of high-surface area electronic materials.

High‐safety and low‐cost aqueous zinc‐ion batteries (ZIBs) are an exceptionally compelling technology for grid‐scale energy storage, whereas the corrosion, hydrogen evolution reaction and dendrites growth of Zn anodes plague their...

No abstract

The dendrite issues associated with zinc anode lead to safety hazards and sluggish reaction kinetics, and largely restrain widespread application of aqueous zinc ion batteries (ZIBs). Herein, a functional separator composed of cellulose nanofibers and graphene oxide (CG) is developed for dendrite‐free and exceptionally stable ZIBs, realized by uniform hexagonal zinc deposits with manipulated crystallographic orientation (002) plane. This CG separator with negative surface charges and abundant zincophilic‐O groups ensures the strong interaction between the separator and zinc species, simultaneously inducing Zn(002) deposition due to the low mismatch between (002)Zn and (002)GO, thus initiating the preferential orientation of the zinc growth along the horizontal direction due to strong Zn binding ability, and uniform interfacial charge of Zn(002) deposition. Furthermore, the CG separator can effectively promote the uniform nucleation of Zn2+ and eliminate side effects. Accordingly, extremely low polarization of 58 mV at 0.5 mA cm−2, and ultralong cycle life over 1750 h at 2 mA cm−2 and 400 h at 20 mA cm−2 are achieved for the zinc anode. Notably, the CG separator significantly boosts rate capability and cyclability of coin‐type full batteries (Zn||Zn(CF3SO3)2||V2O5, Zn||ZnSO4+MnSO4||MnO2/graphite) and a flexible soft‐packaged battery (Zn||MnO2). Therefore, this work introduces a sustainability consideration in to the design of separators for constructing dendrite‐free ZIBs.

Background: Epigenetic alterations are involved in various aspects of colorectal carcinogenesis. N 6methyladenosine (m 6 A) modifications of RNAs are emerging as a new layer of epigenetic regulation. As the most abundant chemical modification of eukaryotic mRNA, m 6 A is essential for the regulation of mRNA stability, splicing, and translation. Alterations of m 6 A regulatory genes play important roles in the pathogenesis of a variety of human diseases. However, whether this mRNA modification participates in the glucose metabolism of colorectal cancer (CRC) remains uncharacterized. Methods: Transcriptome-sequencing and liquid chromatography-tandem mass spectrometry (LC-MS) were performed to evaluate the correlation between m 6 A modifications and glucose metabolism in CRC. Mass spectrometric metabolomics analysis, in vitro and in vivo experiments were conducted to investigate the effects of METTL3 on CRC glycolysis and tumorigenesis. RNA MeRIP-sequencing, immunoprecipitation and RNA stability assay were used to explore the molecular mechanism of METTL3 in CRC. Results: A strong correlation between METTL3 and 18 F-FDG uptake was observed in CRC patients from Xuzhou Central Hospital. METTL3 induced-CRC tumorigenesis depends on cell glycolysis in multiple CRC models. Mechanistically, METTL3 directly interacted with the 5′/3'UTR regions of HK2, and the 3'UTR region of SLC2A1 (GLUT1), then further stabilized these two genes and activated the glycolysis pathway. M 6 A-mediated HK2 and SLC2A1 (GLUT1) stabilization relied on the m 6 A reader IGF2BP2 or IGF2BP2/3, respectively. Conclusions: METTL3 is a functional and clinical oncogene in CRC. METTL3 stabilizes HK2 and SLC2A1 (GLUT1) expression in CRC through an m 6 A-IGF2BP2/3-dependent mechanism. Targeting METTL3 and its pathway offer alternative rational therapeutic targets in CRC patients with high glucose metabolism.

Person re-identification (Re-ID) has achieved great improvement with deep learning and a large amount of labelled training data. However, it remains a challenging task for adapting a model trained in a source domain of labelled data to a target domain of only unlabelled data available. In this work, we develop a self-training method with progressive augmentation framework (PAST) to promote the model performance progressively on the target dataset. Specially, our PAST framework consists of two stages, namely, conservative stage and promoting stage. The conservative stage captures the local structure of target-domain data points with triplet-based loss functions, leading to improved feature representations. The promoting stage continuously optimizes the network by appending a changeable classification layer to the last layer of the model, enabling the use of global information about the data distribution. Importantly, we propose a new self-training strategy that progressively augments the model capability by adopting conservative and promoting stages alternately. Furthermore, to improve the reliability of selected triplet samples, we introduce a ranking-based triplet loss in the conservative stage, which is a label-free objective function basing on the similarities between data pairs. Experiments demonstrate that the proposed method achieves state-of-the-art person Re-ID performance under the unsupervised cross-domain setting.