摘要 :
Zero-shot detection (ZSD) is crucial to large-scale object detection with the aim of simultaneously localizing and recognizing unseen objects- There remain several challenges for ZSD, including reducing the ambiguity between backg...
展开
Zero-shot detection (ZSD) is crucial to large-scale object detection with the aim of simultaneously localizing and recognizing unseen objects- There remain several challenges for ZSD, including reducing the ambiguity between background and unseen objects as well as improving the alignment between visual and semantic concept. In this work, we propose a novel framework named Background Learnable Cascade (BLC) to improve ZSD performance. The major contributions for BLC are as follows: (ⅰ) we propose a multi-stage cascade structure named Cascade Semantic R-CNN to progressively refine the alignment between visual and semantic of ZSD; (ⅱ) we develop the semantic information flow structure and directly add it between each stage in Cascade Semantic R-CNN to further improve the semantic feature learning; (ⅲ) we propose the background learnable region proposal network (BLRPN) to learn an appropriate word vector for background class and use this learned vector in Cascade Semantic R-CNN, this design makes "Background Learnable" and reduces the confusion between background and unseen classes. Our extensive experiments show BLC obtains significantly performance improvements for MS-COCO over state-of-the-art methods.
收起
摘要 :
Zero-shot detection (ZSD) is crucial to large-scale object detection with the aim of simultaneously localizing and recognizing unseen objects- There remain several challenges for ZSD, including reducing the ambiguity between backg...
展开
Zero-shot detection (ZSD) is crucial to large-scale object detection with the aim of simultaneously localizing and recognizing unseen objects- There remain several challenges for ZSD, including reducing the ambiguity between background and unseen objects as well as improving the alignment between visual and semantic concept. In this work, we propose a novel framework named Background Learnable Cascade (BLC) to improve ZSD performance. The major contributions for BLC are as follows: (ⅰ) we propose a multi-stage cascade structure named Cascade Semantic R-CNN to progressively refine the alignment between visual and semantic of ZSD; (ⅱ) we develop the semantic information flow structure and directly add it between each stage in Cascade Semantic R-CNN to further improve the semantic feature learning; (ⅲ) we propose the background learnable region proposal network (BLRPN) to learn an appropriate word vector for background class and use this learned vector in Cascade Semantic R-CNN, this design makes "Background Learnable" and reduces the confusion between background and unseen classes. Our extensive experiments show BLC obtains significantly performance improvements for MS-COCO over state-of-the-art methods.
收起
摘要 :
This paper explores a model of experience-centered service design thinking based on real practice in order to provide practical solutions for systematic optimization in the New Retail Era in China. Service design consists of two p...
展开
This paper explores a model of experience-centered service design thinking based on real practice in order to provide practical solutions for systematic optimization in the New Retail Era in China. Service design consists of two parts, the understanding of service processes and the understanding of target users, and we proposed a five-stages thinking model to better summarize our experiences of how to organize the processes. We designed a flower purchase APP named Picky and applied it to an ordinary community in Haidian District, Beijing. HC1 methods and the re-optimizing of the supply chain greatly improve the efficiency of traditional services, reduce the costs and improve the user experience. Within two months, we have collected 1200 users in this community and beyond. The experiences can be applied to strategically launch new services or optimize traditional services, and it will also be beneficial in establishing platform credibility and effective service mechanisms.
收起
摘要 :
This paper explores a model of experience-centered service design thinking based on real practice in order to provide practical solutions for systematic optimization in the New Retail Era in China. Service design consists of two p...
展开
This paper explores a model of experience-centered service design thinking based on real practice in order to provide practical solutions for systematic optimization in the New Retail Era in China. Service design consists of two parts, the understanding of service processes and the understanding of target users, and we proposed a five-stages thinking model to better summarize our experiences of how to organize the processes. We designed a flower purchase APP named Picky and applied it to an ordinary community in Haidian District, Beijing. HC1 methods and the re-optimizing of the supply chain greatly improve the efficiency of traditional services, reduce the costs and improve the user experience. Within two months, we have collected 1200 users in this community and beyond. The experiences can be applied to strategically launch new services or optimize traditional services, and it will also be beneficial in establishing platform credibility and effective service mechanisms.
收起
摘要 :
1.Both direct H_3PO_4 loss and combined (HPO_3+H_2O) losses contribute to the loss of 98 Da in CID-MS/MS. 2.Combined (HPO_3+H_2O) losses are more prevalent under non-mobile peptide protonation conditions. 3.pT containing peptides ...
展开
1.Both direct H_3PO_4 loss and combined (HPO_3+H_2O) losses contribute to the loss of 98 Da in CID-MS/MS. 2.Combined (HPO_3+H_2O) losses are more prevalent under non-mobile peptide protonation conditions. 3.pT containing peptides produce more combined (HPO_3+H_2O) losses than pS. 4.Carboxylic groups are not a major source of water loss in the combined (HPO_3+H_2O) losses. 5.HCD-MS/MS generates less (HPO_3+H_2O) losses than CID-MS/MS for the same phosphopeptide precursor ion.
收起
摘要 :
1.Both direct H_3PO_4 loss and combined (HPO_3+H_2O) losses contribute to the loss of 98 Da in CID-MS/MS. 2.Combined (HPO_3+H_2O) losses are more prevalent under non-mobile peptide protonation conditions. 3.pT containing peptides ...
展开
1.Both direct H_3PO_4 loss and combined (HPO_3+H_2O) losses contribute to the loss of 98 Da in CID-MS/MS. 2.Combined (HPO_3+H_2O) losses are more prevalent under non-mobile peptide protonation conditions. 3.pT containing peptides produce more combined (HPO_3+H_2O) losses than pS. 4.Carboxylic groups are not a major source of water loss in the combined (HPO_3+H_2O) losses. 5.HCD-MS/MS generates less (HPO_3+H_2O) losses than CID-MS/MS for the same phosphopeptide precursor ion.
收起
摘要 :
The thermal stability, oxygen depletion and tensile properties of low density polyethylene (LDPE) resins filled with ascorbic acid (Vc), sodium ascorbate (SA), iron (Fe) and modified iron (MFe) oxygen scavengers were systematicall...
展开
The thermal stability, oxygen depletion and tensile properties of low density polyethylene (LDPE) resins filled with ascorbic acid (Vc), sodium ascorbate (SA), iron (Fe) and modified iron (MFe) oxygen scavengers were systematically investigated. Thermogravimetric analysis (TGA) results clearly suggest that the thermal stability of SA powder and L_(95)(SA)_5 specimen is significantly better than that of Vc powder and L_(95)(Vc)_5 specimen, respectively. The oxygen depletion efficiency of L_(95)(SA)_5 is significantly better than that of L_(95)(Vc)_5, L_(95)(Fe)_5 and L_(95)( MFe)_5 specimens, although the virgin SA powders exhibit worse oxygen depletion efficiency than Vc, Fe or MFe powders before melt blending. Moreover, at a fixed weight ratio of Vc (or SA) to MFe of the oxygen scavenger compounds, the oxygen depletion efficiency of L95[SAx(MFe)y]5 series specimens is always significantly better than that of L95[Vcx(MFe)y]5 series specimens. In fact, at weight ratios of Vc/MFe and SA/MFe higher than 3/7 and 5/5, respectively, the residual oxygen concentration values present in the airtight flask of L95[Vcx(MFe)y]5 and L95[SAx( MFe)y]5 series samples at any time are even lower than those of the L95(Vc)5 and L95(SA)5 specimens, respectively. Further tensile experiments show that the tensile properties of the L95[SAx(MFe)y]5 series samples are always higher than those of the corresponding L95[Vcx(MFe)y]5 series samples with the same loadings of oxygen scavenger compounds, respectively. In order to understand these interesting thermal stability, oxygen depletion and tensile properties of these LDPE oxygen-scavenging plastics, scanning electron microscope and energy dispersive X-rays analysis of the compositions on the surfaces of L95[SAx(MFe)y]5 and L95[Vcx(MFe)y]5 series samples were performed. Possible reasons accounting for these interesting properties of these LDPE oxygen-scavenging plastics are proposed.
收起
摘要 :
Sequential recommendation aims to suggest items to users based on sequential dependencies. Graph neural networks (GNNs) are recently proposed to capture transitions of items by treating session sequences as graph-structured data. ...
展开
Sequential recommendation aims to suggest items to users based on sequential dependencies. Graph neural networks (GNNs) are recently proposed to capture transitions of items by treating session sequences as graph-structured data. However, existing graph construction approaches mainly focus on the directional dependency of items and ignore benefits of feature aggregation from undirectional relationship. In this paper, we innovatively propose a joint graph contextualized network (JGCN) for sequential recommendation, which constructs both the directed graphs and undirected graphs to jointly capture current interests and global preferences. Specifically, we introduce gate graph neural networks and model the combined embedding of weighted position and node information from directed graphs for capturing current interests. Besides, to learn global preferences, we propose a graph collaborative attention network with correlation-based similarity of items from undirected graphs. Finally, a feed-forward layer with the residual connection is applied to synthetically obtain accurate transitions of items. Extensive experiments conducted on three datasets show that JGCN outperforms state-of-the-art methods.
收起
摘要 :
Sequential recommendation aims to suggest items to users based on sequential dependencies. Graph neural networks (GNNs) are recently proposed to capture transitions of items by treating session sequences as graph-structured data. ...
展开
Sequential recommendation aims to suggest items to users based on sequential dependencies. Graph neural networks (GNNs) are recently proposed to capture transitions of items by treating session sequences as graph-structured data. However, existing graph construction approaches mainly focus on the directional dependency of items and ignore benefits of feature aggregation from undirectional relationship. In this paper, we innovatively propose a joint graph contextualized network (JGCN) for sequential recommendation, which constructs both the directed graphs and undirected graphs to jointly capture current interests and global preferences. Specifically, we introduce gate graph neural networks and model the combined embedding of weighted position and node information from directed graphs for capturing current interests. Besides, to learn global preferences, we propose a graph collaborative attention network with correlation-based similarity of items from undirected graphs. Finally, a feed-forward layer with the residual connection is applied to synthetically obtain accurate transitions of items. Extensive experiments conducted on three datasets show that JGCN outperforms state-of-the-art methods.
收起
摘要 :
Carbon nanotubes (CNTs) demonstrate unique mechanical properties with extraordinarily high stiffness, strength and resilience that make them as an ideal reinforcing material for developing new nanocomposites. Due to the current CN...
展开
Carbon nanotubes (CNTs) demonstrate unique mechanical properties with extraordinarily high stiffness, strength and resilience that make them as an ideal reinforcing material for developing new nanocomposites. Due to the current CNT technology limitations, it is difficult to prepare and design new nanocomposites based on experiment approaches which are both expensive and time-consuming. Computational simulation has been recognized as one of powerful tools in overcoming these problems. In this paper, the authors will utilize the multi-scale modeling method to study the bending characteristics of CNT and CNT- reinforced composites. Firstly, a combining molecular dynamics and continuum mechanics based model will be applied on the CNTs, and their tensional and flexural modulus will be obtained through tension and bending simulation analysis. Then the CNT model is simplified to several 3D beams and is inserted into the matrix, thus, a multi-scale RVE (Representative volume element) model of CNT-reinforced composite is established. Using this RVE model, the bending characteristics of CNT-based composites are obtained. The influence of diameter D, length L, aspect ratio L/D, volume fraction, chiral of CNT and shape of RVE as well as the arrangement of CNT in matrix on the reinforcement effect of flexural modulus of resultant nanocomposites will be further discussed.
收起
原文获取