Filters








12,366 Hits in 4.8 sec

Discriminative Distance-Based Network Indices with Application to Link Prediction [article]

Mostafa Haghir Chehreghani, Albert Bifet, Talel Abdessalem
2018 arXiv   pre-print
Then, we show that real-world networks have usually a tiny average discriminative path length, bounded by a constant (e.g., 2).  ...  In this paper, first we develop a new distance measure between vertices of a graph that yields discriminative distance-based centrality indices.  ...  We call this property the tiny-world property.  ... 
arXiv:1703.06227v3 fatcat:fiajas7f4betzlq2taxel3hree

A Normalized Gaussian Wasserstein Distance for Tiny Object Detection [article]

Jinwang Wang, Chang Xu, Wen Yang, Lei Yu
2021 arXiv   pre-print
Our key observation is that Intersection over Union (IoU) based metrics such as IoU itself and its extensions are very sensitive to the location deviation of the tiny objects, and drastically deteriorate  ...  The proposed NWD metric can be easily embedded into the assignment, non-maximum suppression, and loss function of any anchor-based detector to replace the commonly used IoU metric.  ...  Compared to baseline method, NWD-based assigning module in RPN and R-CNN respectively achieves the highest and second-highest AP improvement of 6.2% and 3.2%, which indicates that the problem of tiny object  ... 
arXiv:2110.13389v1 fatcat:ckbv75ypw5hhpn4cltfb5kdq64

Tiny Object Tracking: A Large-scale Dataset and A Baseline [article]

Yabin Zhu, Chenglong Li, Yao Liu, Xiao Wang, Jin Tang, Bin Luo, Zhixiang Huang
2022 arXiv   pre-print
effectively enhance the feature representation, discrimination and localization abilities in tracking tiny objects.  ...  In data creation, we take 12 challenge attributes into account to cover a broad range of viewpoints and scene complexities, and annotate these attributes for facilitating the attribute-based performance  ...  The third category is trackers based on discriminant feature learning, including ATOM [32] and DiMP [21] .  ... 
arXiv:2202.05659v1 fatcat:64dbxxjrb5fotnl645g3axsi24

Towards Generalized Implementation of Wasserstein Distance in GANs [article]

Minkai Xu, Zhiming Zhou, Guansong Lu, Jian Tang, Weinan Zhang, Yong Yu
2021 arXiv   pre-print
Theoretically, we first demonstrate a more general dual form of the Wasserstein distance called the Sobolev duality, which relaxes the Lipschitz constraint but still maintains the favorable gradient property  ...  Based on the relaxed duality, we further propose a generalized WGAN training scheme named Sobolev Wasserstein GAN (SWGAN), and empirically demonstrate the improvement of SWGAN over existing methods with  ...  The generator network G learns to map samples from a noise distribution to a target distribution, while the discriminator network D is trained to distinguish between the real data and the generated samples  ... 
arXiv:2012.03420v2 fatcat:akewv6hz3jew7lqtnd6sdcvghi

A Data Augmentation Method Based on Generative Adversarial Networks for Grape Leaf Disease Identification

Bin Liu, Cheng Tan, Shuqin Li, Jinrong He, Hongyan Wang
2020 IEEE Access  
The generated grape leaf disease images based on Leaf GAN model can obtain better performance than DCGAN and WGAN in terms of the Fréchet inception distance.  ...  For the eight prevailing classification models with the expanded dataset, the identification performance based on CNNs indicated higher accuracies, whereby all the accuracies were better than those of  ...  The adversarial loss of the discriminator model is defined based on the original GAN model, and the formula is as shown in Eq. (2): L adv = −E x∼P data log D(x) − E z−P G log(1 − D(G(z))) (2) P data indicates  ... 
doi:10.1109/access.2020.2998839 fatcat:3pn7nigtzjcy7h4ueydkvcaqgy

Perceptual Image Super-Resolution with Progressive Adversarial Network [article]

Lone Wong, Deli Zhao, Shaohua Wan, Bo Zhang
2020 arXiv   pre-print
The key principle of PAN is that we do not apply any distance-based reconstruction errors as the loss to be optimized, thus free from the restriction of the curse of dimensionality.  ...  Extensive experiments demonstrate the superiority of our algorithm over state-of-the-arts both quantitatively and qualitatively.  ...  When measuring the distance between super-solved images and ground-truth images, the norm-based distances (e.g. L 1 and L 2 norms) are indispensable for existing algorithmic frameworks.  ... 
arXiv:2003.03756v4 fatcat:dg32vyec5ndhrhmpin7kp4uhwi

WiSH: WiFi-based real-time human detection

Tianmeng Hang, Yue Zheng, Kun Qian, Chenshu Wu, Zheng Yang, Xiancun Zhou, Yunhao Liu, Guilin Chen
2019 Tsinghua Science and Technology  
We deploy WiSH on commodity desktops and customized tiny nodes in different everyday scenarios.  ...  Among numerous applications of WiFi-based sensing, human presence detection acts as a primary and fundamental function to boost applications in practice.  ...  Acknowledgment This work was supported in part by the National Key Research Plan (No. 2016YFC0700100), the National Natural Science Foundation of China (Nos. 61832010, 61332004, and 61572366).  ... 
doi:10.26599/tst.2018.9010091 fatcat:rern4nae3rgf3acbhl7j6wstny

M2GAN: A Multi-Stage Self-Attention Network for Image Rain Removal on Autonomous Vehicles [article]

Duc Manh Nguyen, Sang-Woong Lee
2021 arXiv   pre-print
To demonstrate M2GAN, we introduce the first real-world dataset for rain removal on autonomous vehicles.  ...  The experimental results show that our proposed method is superior to other state-of-the-art approaches of deraining raindrops in respect of quantitative metrics and visual quality.  ...  Extensive experimental results demonstrated that M2GAN performed considerably better than state-of-the-art methods in handling real-world raindrops and rain flows.  ... 
arXiv:2110.06164v1 fatcat:wqb5mq7um5c5tbfycjrhwjspre

Directional Adversarial Training for Cost Sensitive Deep Learning Classification Applications [article]

Matteo Terzi, Gian Antonio Susto, Pratik Chaudhari
2019 arXiv   pre-print
Moreover, WPGD solves an optimal transport problem on the output space of the network and it can efficiently discover directions where robustness is required, allowing to control the directional trade-off  ...  Unfortunately, one of the main issues in adversarial training is that robustness w.r.t. gradient-based attackers is always achieved at the cost of prediction accuracy.  ...  Acknowledgments The authors gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan V GPU used for this research and Amazon Web Services for donating research credits.  ... 
arXiv:1910.03468v1 fatcat:zjnuxjkxfre6fitqueniw6eonu

Graph Self-supervised Learning with Accurate Discrepancy Learning [article]

Dongki Kim, Jinheon Baek, Sung Ju Hwang
2022 arXiv   pre-print
To tackle such limitations, we propose a framework that aims to learn the exact discrepancy between the original and the perturbed graphs, coined as Discrepancy-based Self-supervised LeArning (D-SLA).  ...  Moreover, we further aim to accurately capture the amount of discrepancy for each perturbed graph using the graph edit distance.  ...  Distance-based Discrepancy Learning Based upon the graph edit distance between the original and its perturbed graphs, we now design the regularization term to learn the exact amount of differences between  ... 
arXiv:2202.02989v2 fatcat:7x2xshtofvbjlobnqzngkxkosm

An Efficient Multilevel Probabilistic Model for Abnormal Traffic Detection in Wireless Sensor Networks

Muhammad Altaf Khan, Moustafa M. Nasralla, Muhammad Muneer Umar, Ghani-Ur-Rehman, Shafiullah Khan, Nikumani Choudhury
2022 Sensors  
This paper introduces a novel mechanism based on a Bayesian model to detect abnormal data traffic and discriminate DDoS attacks from FC in it.  ...  Wireless sensor networks (WSNs) are low-cost, special-purpose networks introduced to resolve various daily life domestic, industrial, and strategic problems.  ...  Acknowledgments: The authors would like to acknowledge Prince Sultan University (PSU) and Smart Systems Engineering Lab for their valuable support and provision of research facilities essential for completing  ... 
doi:10.3390/s22020410 pmid:35062372 pmcid:PMC8777834 fatcat:cm75rfa23rea5bohwj7dsze7uy

Deep Comprehensive Correlation Mining for Image Clustering [article]

Jianlong Wu, Keyu Long, Fei Wang, Chen Qian, Cheng Li, Zhouchen Lin, Hongbin Zha
2019 arXiv   pre-print
to image transformation of input space is fully explored, which benefits the network learning and significantly improves the performance. 3) The triplet mutual information among features is presented  ...  data from three aspects: 1) Instead of only using pair-wise information, pseudo-label supervision is proposed to investigate category information and learn discriminative features. 2) The features' robustness  ...  It has the following properties: K t=1 z it = 1, ∀i = 1, · · · , N, and z it ≥ 0, ∀t = 1, · · · , K. (1) Based on the label feature z, the cosine similarity between the i-th and the j-th samples can be  ... 
arXiv:1904.06925v3 fatcat:ljw43lvlcbcxxetf7wfgpupqfe

DOCTOR: A Simple Method for Detecting Misclassification Errors [article]

Federica Granese, Marco Romanelli, Daniele Gorla, Catuscia Palamidessi, Pablo Piantanida
2021 arXiv   pre-print
Deep neural networks (DNNs) have shown to perform very well on large scale object recognition problems and lead to widespread use for real-world applications, including situations where DNN are implemented  ...  DOCTOR can be applied to any pre-trained model, it does not require prior information about the underlying dataset and is as simple as the simplest available methods in the literature.  ...  Clearly, the Mahalanobis distance-based method requires additional samples compared to DOCTOR.  ... 
arXiv:2106.02395v2 fatcat:zxgpguawkfh6nosgghy6hd7n4m

Deep Rival Penalized Competitive Learning for Low-resolution Face Recognition [article]

Peiying Li, Shikui Tu, Lei Xu
2021 arXiv   pre-print
which enforces intra-class compactness and inter-class discrepancy.  ...  Inspired by the idea of the RPCL, our method further enforces regulation on the rival logit, which is defined as the largest non-target logit for an input image.  ...  Based on this property, we present a deep RPCL learning for face recognition.  ... 
arXiv:2108.01286v1 fatcat:tf56qav62zbhjcfg2ywobvkpeu

A one-pass resource-allocating codebook for patch-based visual object recognition

Amirthalingam Ramanan, Mahesan Niranjan
2010 2010 IEEE International Workshop on Machine Learning for Signal Processing  
It simultaneously achieves increased discrimination and a drastic reduction in the computational needs. We illustrate some properties of our method and compare it to a closely related approach.  ...  However, clustering is a process that retains regions of high density in a distribution and it follows that the resulting codebook need not have discriminant properties.  ...  Acknowledgment AR is supported in parts by a grant from the University of Jaffna, Sri Lanka and the University of Southampton, UK.  ... 
doi:10.1109/mlsp.2010.5589204 fatcat:uvja5o5yufhkniqoy2icfyaeiu
« Previous Showing results 1 — 15 out of 12,366 results