text
stringlengths
27
153
label
stringlengths
27
153
id
int64
0
40k
large an different However, is there number practical of models in now use. astonishingly
However, there is now an astonishingly large number of different models in practical use.
38,600
The semantic trackers. of Siamese as distractors, are the considered robustness which backgrounds hinders always
The semantic backgrounds are always considered as distractors, which hinders the robustness of Siamese trackers.
38,601
long-term we tracking. learning this focus and Siamese networks accurate In distractor-aware for paper, on
In this paper, we focus on learning distractor-aware Siamese networks for accurate and long-term tracking.
38,602
in at are end, traditional Siamese trackers this features used first. analyzed To
To this end, features used in traditional Siamese trackers are analyzed at first.
38,603
the learned features makes imbalanced data observe the less distribution that of training discriminative. We
We observe that the imbalanced distribution of training data makes the learned features less discriminative.
38,604
asset digital in are management power for development system. grid need Chinese of enterprises
Chinese power grid enterprises are in need for development of digital asset management system.
38,605
the have training the a priori all in known However, operations to be phase.
However, all the operations have to be known a priori in the training phase.
38,606
transit consumers rail large Electric systems are energy. of
Electric rail transit systems are large consumers of energy.
38,607
substations. reversible and technologies storage systems optimization, (onboard include: and timetable energy These train wayside),
These technologies include: train timetable optimization, energy storage systems (onboard and wayside), and reversible substations.
38,608
generation. approach uses planning It to a story
It uses a planning approach to story generation.
38,609
pre- the the raw sensor need process data. eliminates or The post network to
The network eliminates the need to pre- or post process the raw sensor data.
38,610
great supervised unsupervised challenges. and methods face Existing
Existing supervised and unsupervised methods face great challenges.
38,611
this novel data be Cross-domain in utilized framework. synthetic fully could
Cross-domain synthetic data could be fully utilized in this novel framework.
38,612
ensure well are transferred perception capability domains. strategies across to depth proposed Different different learned
Different strategies are proposed to ensure learned depth perception capability well transferred across different domains.
38,613
dataset. show results depth state-of-the-art of on KITTI extensive experiments Our monocular estimation
Our extensive experiments show state-of-the-art results of monocular depth estimation on KITTI dataset.
38,614
architecture. connectivity U-Net a We the pattern design for new
We design a new connectivity pattern for the U-Net architecture.
38,615
the connections flow across efficiently information could U-Nets. coupling The more make
The coupling connections could make the information flow more efficiently across U-Nets.
38,616
parameter feature makes across reuse U-Net each very U-Nets The efficient.
The feature reuse across U-Nets makes each U-Net very parameter efficient.
38,617
estimation. evaluate U-Nets of datasets pose the coupled human on benchmark We two
We evaluate the coupled U-Nets on two benchmark datasets of human pose estimation.
38,618
are number accuracy parameter and model compared. the Both
Both the accuracy and model parameter number are compared.
38,619
accuracy CU-Net comparable obtains The state-of-the-art as methods.
The CU-Net obtains comparable accuracy as state-of-the-art methods.
38,620
the retrospective most not existing datasets consist of sequences. FLAIR do However, of
However, most of the existing retrospective datasets do not consist of FLAIR sequences.
38,621
process methods the and process imputation, of modality the segmentation. of imputation Existing missing separate
Existing missing modality imputation methods separate the process of imputation, and the process of segmentation.
38,622
are segmentation. Iso-depth via regions determined superpixel
Iso-depth regions are determined via superpixel segmentation.
38,623
model. a dissertation, probabilistic two index solutions provide we In this to fit
In this dissertation, we provide two solutions to fit a probabilistic index model.
38,624
entire unique into of set partitions. data consists splitting the The first algorithm
The first algorithm consists of splitting the entire data set into unique partitions.
38,625
fit the aggregate model these, then On each of and we the estimates.
On each of these, we fit the model and then aggregate the estimates.
38,626
we in the work, this address variation. weather of that is One challenges, particular
One of the particular challenges, that we address in this work, is weather variation.
38,627
low-level can not model properly. do produce that drastic Seasonal changes, features classic changes appearance
Seasonal changes can produce drastic appearance changes, that classic low-level features do not model properly.
38,628
Our contributions paper in are twofold. this
Our contributions in this paper are twofold.
38,629
the outperform the best Our art of field. the of results state
Our best results outperform the state of the art of the field.
38,630
procedure, time-consuming and error-prone. This is however,
This procedure, however, is time-consuming and error-prone.
38,631
Using mitosis ensembling combined detector. stain invariant with a network it during in resulted training
Using it during training combined with network ensembling resulted in a stain invariant mitosis detector.
38,632
issues two metric important are Feature person for and learning re-identification. representation
Feature representation and metric learning are two important issues for person re-identification.
38,633
applications technology medium can find in link power reconfigurable ac-dc distribution. interesting Parallel voltage
Parallel ac-dc reconfigurable link technology can find interesting applications in medium voltage power distribution.
38,634
suitable time makes applications. it for The of fast model real-time execution the
The fast execution time of the model makes it suitable for real-time applications.
38,635
overcome. nanoscale Powering with challenge major such wireless communications is supply, however, a energy to
Powering such wireless communications with nanoscale energy supply, however, is a major challenge to overcome.
38,636
text-to-image heterogeneous synthesis in homogeneous lie main of and two The gaps: issues the gaps.
The main issues of text-to-image synthesis lie in two gaps: the heterogeneous and homogeneous gaps.
38,637
discriminative exploit these of we (e.g. problems, generic addressing models excellent capability the For
For addressing these problems, we exploit the excellent capability of generic discriminative models (e.g.
38,638
datasets two Experiments conducted our are the to verify SDN. proposed of widely-used on effectiveness
Experiments on two widely-used datasets are conducted to verify the effectiveness of our proposed SDN.
38,639
This them applications. makes for preferable real-world
This makes them preferable for real-world applications.
38,640
training require datasets. large networks such However, very
However, such networks require very large training datasets.
38,641
laborious. difficult is data training Engineering the and/or
Engineering the training data is difficult and/or laborious.
38,642
As on results the obtain consequence, KITTI we benchmarks. a state-of-the-art
As a consequence, we obtain state-of-the-art results on the KITTI benchmarks.
38,643
code manuscript. this The matlab available is in
The matlab code is available in this manuscript.
38,644
any bias. in and terms we As for logic approximation, efficiency subjective computational evaluate of
As for any approximation, we evaluate subjective logic in terms of computational efficiency and bias.
38,645
IHC to co-registered H&E slides the were The slides.
The H&E slides were co-registered to the IHC slides.
38,646
imagery airplane a Automatic in detection variety of applications. aerial has
Automatic airplane detection in aerial imagery has a variety of applications.
38,647
rotation-and-scale a solve invariant generator. present challenges, these we proposal airplane To
To solve these challenges, we present a rotation-and-scale invariant airplane proposal generator.
38,648
non-maximum suppression remove to detections. Finally, employ we duplicate
Finally, we employ non-maximum suppression to remove duplicate detections.
38,649
annotations as Moreover, an of airplanes box-level using estimate direction the we achievement. the extra
Moreover, we estimate the direction of the airplanes using box-level annotations as an extra achievement.
38,650
We images. dermoscopic lesion strategy for superpixel-based on present skin segmenting a
We present a superpixel-based strategy for segmenting skin lesion on dermoscopic images.
38,651
as mean superpixel RGB The used color merging criterion. of was each
The mean RGB color of each superpixel was used as merging criterion.
38,652
the a framework realize is the challenge. in multi-view fusion How such primary to joint
How to realize the multi-view fusion in such a joint framework is the primary challenge.
38,653
clustering DMJC-T KL are objective. a Both under optimized and DMJC-S like divergence
Both DMJC-S and DMJC-T are optimized under a KL divergence like clustering objective.
38,654
learn network in an rectification We the network and end-to-end recognition manner.
We learn the rectification network and recognition network in an end-to-end manner.
38,655
graph. Our model consists one four tensorflow final aggregated models single into of
Our final model consists of four single models aggregated into one tensorflow graph.
38,656
a problem. two-stage this tacking We deep for framework learning present
We present a two-stage deep learning framework for tacking this problem.
38,657
new such and image Our portrait extrapolation. removal portrait editing applications method enables as occlusion
Our method enables new portrait image editing applications such as occlusion removal and portrait extrapolation.
38,658
several performance datasets. state-of-the-art on in results a This
This results in a state-of-the-art performance on several datasets.
38,659
The knowledge representation ASP presents a extends language paper with aggregates. $\mathcal{A}log$ which
The paper presents a knowledge representation language $\mathcal{A}log$ which extends ASP with aggregates.
38,660
capture high imaging frame to ultrasound Cardiac rapid motion. requires order a in rate
Cardiac ultrasound imaging requires a high frame rate in order to capture rapid motion.
38,661
expense acquisition introducing block This shortens artifacts. the of at time the
This shortens the acquisition time at the expense of introducing block artifacts.
38,662
improve quality. MLA we the propose In data-driven image this learning-based to a paper, approach
In this paper, we propose a data-driven learning-based approach to improve the MLA image quality.
38,663
endoscopy capsule diagnosis Wireless (WCE) gastrointestinal mean for effective an disorders. of is
Wireless capsule endoscopy (WCE) is an effective mean for diagnosis of gastrointestinal disorders.
38,664
of the simplified Both MLP and computational CNN number reduce are operations. to structures
Both CNN and MLP structures are simplified to reduce the number of computational operations.
38,665
convolutional regularization spatial for strategies neural (e.g. with networks)
convolutional neural networks) with strategies for spatial regularization (e.g.
38,666
models conditional as such fields). graphical random
graphical models such as conditional random fields).
38,667
dominate the to started re-identification Deep video-based progress have methods person learning (re-id). research of
Deep learning methods have started to dominate the research progress of video-based person re-identification (re-id).
38,668
lack and real-world scalability video surveillance Therefore, they in severely practicality applications.
Therefore, they severely lack scalability and practicality in real-world video surveillance applications.
38,669
employed can scheme. our readily standard CNNs DAL Existing be within
Existing standard CNNs can be readily employed within our DAL scheme.
38,670
promising in Convolutional Recently, Neural have (SR). super-resolution performance (CNNs) Networks shown
Recently, Convolutional Neural Networks (CNNs) have shown promising performance in super-resolution (SR).
38,671
branch. IRL, as master pre-trained first typical select we a In SR network a
In IRL, first we select a typical SR pre-trained network as a master branch.
38,672
comprehensive, for a for Finally, implications cost-effective malaria discuss strategy we the control.
Finally, we discuss the implications for a comprehensive, cost-effective strategy for malaria control.
38,673
It component software critical based has and systems. is vision applications spread for wide a
It has wide spread applications and is a critical component for vision based software systems.
38,674
plane causes in and the printed phenomenon, misregistration, image. halo gap color artifacts This called
This phenomenon, called color plane misregistration, causes gap and halo artifacts in the printed image.
38,675
algorithms embedded are for Our or designed firmware implementation. software
Our algorithms are designed for software or embedded firmware implementation.
38,676
based the are first of use The algorithms two look-up tables on (LUTs).
The first two algorithms are based on the use of look-up tables (LUTs).
38,677
in firmware particularly printers. low-cost the attractive of for algorithm embedded This is formatter-based implementation
This algorithm is particularly attractive for implementation in the embedded firmware of low-cost formatter-based printers.
38,678
business. is dataset applied model open real the to an for And
And the model is applied to an open dataset for real business.
38,679
without with any labelled data. supervised been facial attributes, having
facial attributes, without having been supervised with any labelled data.
38,680
index data of pseudo-Rsquared proposed. to new is goodness-of-fit VBGF A the models LFD for
A new pseudo-Rsquared index for the goodness-of-fit of VBGF models to LFD data is proposed.
38,681
curve and robust, bootstrap-based are for presented New, fitting methods discussed.
New, robust, bootstrap-based methods for curve fitting are presented and discussed.
38,682
regularized proposed interval-valued for neural (RANN) artificial is prediction. network A data
A regularized artificial neural network (RANN) is proposed for interval-valued data prediction.
38,683
in to introduce reduce paper, method we arising MLT. this In artifacts a data-driven the
In this paper, we introduce a data-driven method to reduce the artifacts arising in MLT.
38,684
lane-level accuracy for autonomous localization is very vehicles. important The
The lane-level localization accuracy is very important for autonomous vehicles.
38,685
Satellite The e.g. Global (GNSS), System Navigation
The Global Navigation Satellite System (GNSS), e.g.
38,686
classification with was compared the method. The SVM performance CNN the of based method
The performance of the SVM based classification method was compared with the CNN method.
38,687
from SFP scratch prune enables capacity the and Large model train to simultaneously.
Large capacity enables SFP to train from scratch and prune the model simultaneously.
38,688
pruning outperforms filter scratch SFP Empirically, the previous methods. from
Empirically, SFP from scratch outperforms the previous filter pruning methods.
38,689
effective CNN many Moreover, advanced for approach demonstrated has architectures. been our
Moreover, our approach has been demonstrated effective for many advanced CNN architectures.
38,690
separates This from avoids recovering high frequencies, denoising over-smoothing. which features low-frequency
This separates recovering low-frequency features from denoising high frequencies, which avoids over-smoothing.
38,691
signed (inside/outside) The distances the samples. output balances to the curve
The output curve balances the signed distances (inside/outside) to the samples.
38,692
connectivity. on parameter-free is Our fast determined the algorithm local the neighborhoods operates by and
Our algorithm is parameter-free and operates fast on the local neighborhoods determined by the connectivity.
38,693
solver system by handle to bounds. We augment a least-squares linear a also constrained
We augment a least-squares solver constrained by a linear system to also handle bounds.
38,694
samplers MCMC Carlo computation. an (SMC) attractive Sequential Bayesian for form alternative Monte to
Sequential Monte Carlo (SMC) samplers form an attractive alternative to MCMC for Bayesian computation.
38,695
on rejuvenate used their kernels strongly depends Markov particles. performance to However, the
However, their performance depends strongly on the Markov kernels used to rejuvenate particles.
38,696
Learning problem data important vision. transformation an is visual representations of computer in invariant
Learning transformation invariant representations of visual data is an important problem in computer vision.
38,697
classification for results demonstrated Deep networks have and image tasks. convolutional video remarkable
Deep convolutional networks have demonstrated remarkable results for image and video classification tasks.
38,698
computation Neural achieve superior performance wider bring Deeper (CNNs) and expensive Convolutional but Networks cost.
Deeper and wider Convolutional Neural Networks (CNNs) achieve superior performance but bring expensive computation cost.
38,699