text
stringlengths
27
153
label
stringlengths
27
153
id
int64
0
40k
well Our even hand, method, other datasets. transferred when performs on between the
Our method, on the other hand, performs well even when transferred between datasets.
37,900
The estimate to task between visual person person images. re-identification similarities robustly requires
The person re-identification task requires to robustly estimate visual similarities between person images.
37,901
not similarity be hard of samples the estimation a As some accurate. might result,
As a result, the similarity estimation of some hard samples might not be accurate.
37,902
relation features for similarity prediction. such be achieved probe-gallery estimation can by using updated Accurate
Accurate similarity estimation can be achieved by using such updated probe-gallery relation features for prediction.
37,903
method validated The datasets. person effectiveness three on re-identification proposed public of our is
The effectiveness of our proposed method is validated on three public person re-identification datasets.
37,904
is in people Automated counting of task. crowd a challenging images
Automated counting of people in crowd images is a challenging task.
37,905
large in in appear major diversity the the difficulty from The stems crowds. people way
The major difficulty stems from the large diversity in the way people appear in crowds.
37,906
creating experts recursively, two every networks The split at are again division. child
The child networks are again split recursively, creating two experts at every division.
37,907
major count model datasets. achieves on higher proposed The crowd accuracy
The proposed model achieves higher count accuracy on major crowd datasets.
37,908
the we analyse characteristics mined Further, specialties automatically our method. by of
Further, we analyse the characteristics of specialties mined automatically by our method.
37,909
Electric is electric energy important meter in An component an power service.
An Electric power meter is an important component in electric energy service.
37,910
complained have meter. about the reading the In past, inaccurate consumers electric of many
In the past, many consumers have complained about reading inaccurate of the electric meter.
37,911
of an This RFID development electrical reader. presents with research power the meter equipped
This research presents the development of an electrical power meter equipped with RFID reader.
37,912
dominated are RCNN. region-based segmentation by such Object as and detection instance methods Mask
Object detection and instance segmentation are dominated by region-based methods such as Mask RCNN.
37,913
a the We rate real propose world detection improves in conditions. algorithm new which
We propose a new algorithm which improves the detection rate in real world conditions.
37,914
which has scheme speed. been added processing the tracking also A simple improves
A simple tracking scheme has also been added which improves the processing speed.
37,915
proposed the algorithms outperformed performance. of real-time while the achieving state art algorithm The
The proposed algorithm outperformed the state of the art algorithms while achieving real-time performance.
37,916
(MAC) input. are billions operations intensive, compute multiply-accumulate CNNs of per requiring However,
However, CNNs are compute intensive, requiring billions of multiply-accumulate (MAC) operations per input.
37,917
the inherent damages This correlations among the data. consequently
This consequently damages the inherent correlations among the data.
37,918
for affine Besides, is used practical enough not the applications. methods these in constraint flexible
Besides, the affine constraint used in these methods is not flexible enough for practical applications.
37,919
literature We to references code. also relevant of provide and the most
We also provide references to most of the relevant literature and code.
37,920
computational solutions their in presented complexity. performance and vary The
The solutions presented vary in their performance and computational complexity.
37,921
performance simulations. is these compared of The evaluated via solutions and
The performance of these solutions is evaluated and compared via simulations.
37,922
work, sharp In make post-processing. boundaries is this produce to the without aim CNNs
In this work, the aim is to make CNNs produce sharp boundaries without post-processing.
37,923
which tackle bottom-up/top-down to an the adopts network we architecture propose end-to-end task. the Moreover,
Moreover, we propose an end-to-end network which adopts the bottom-up/top-down architecture to tackle the task.
37,924
We results. following the on our find based closed-form solution
We find the closed-form solution based on our following results.
37,925
same the The is neighboring eigenfunction points. the a of sign on low-frequency
The sign of a low-frequency eigenfunction is the same on the neighboring points.
37,926
knowledge natural in discover trained further The networks scenes. visual are to applied
The trained networks are further applied to discover visual knowledge in natural scenes.
37,927
model then can and lightweight, sophisticated be This for editing. video used image
This model can then be used for lightweight, sophisticated video and image editing.
37,928
our editing. for video We of show using framework also face examples
We also show examples of using our framework for video face editing.
37,929
each are the by corresponding availability such limited scans of However, modality. of techniques
However, such techniques are limited by the availability of corresponding scans of each modality.
37,930
method classification. evaluated Alzheimer's of of efficacy is the in disease The context proposed the
The efficacy of the proposed method is evaluated in the context of Alzheimer's disease classification.
37,931
features residual attenuation. First, without is to network employed hierarchical explore adapted an
First, an adapted residual network is employed to explore hierarchical features without attenuation.
37,932
usually works Previous video clips. video captioning on focus on
Previous works on video captioning usually focus on video clips.
37,933
regions. some The Hungarian algorithm matching performed unmatched is permitting via
The matching is performed via Hungarian algorithm permitting some unmatched regions.
37,934
The similarity context is proposed a clustering the set of employed shapes. of measure in
The proposed similarity measure is employed in the context of clustering a set of shapes.
37,935
degradation common a is source. Image blur
Image blur is a common degradation source.
37,936
motion presence the In with of this focus work, super-resolution task the on we blur.
In this work, we focus on the super-resolution task with the presence of motion blur.
37,937
very role image deblurring applications. multimedia important and Blind vision a plays many in
Blind image deblurring plays a very important role in many vision and multimedia applications.
37,938
is established framework In the above to work, issues. a this address collaborative learning
In this work, a collaborative learning framework is established to address the above issues.
37,939
to similarities person images. robustly between measure re-identification aims Person
Person re-identification aims to robustly measure similarities between person images.
37,940
variation The significant challenges poses and accurate viewing of re-identification. person angles person for
The significant variation of person poses and viewing angles challenges for accurate person re-identification.
37,941
depends dramatic ads climax particular, In changes sentiment. of in on structure the in
In the dramatic structure of ads in particular, climax depends on changes in sentiment.
37,942
to of structure ads study video the dynamic We automatically. dedicate our understand
We dedicate our study to understand the dynamic structure of video ads automatically.
37,943
use supervised to unsupervised both and predict We then methods the climax.
We then use both unsupervised and supervised methods to predict the climax.
37,944
complex has problem six solutions (including We that the generically show ones).
We show that the problem generically has six solutions (including complex ones).
37,945
stable that experiments correct, is numerically robust. demonstrate and method The the
The experiments demonstrate that the method is correct, numerically stable and robust.
37,946
modalities a different from task. challenging Learning is
Learning from different modalities is a challenging task.
37,947
person. have exaggerations a facial features of of Caricature
Caricature have exaggerations of facial features of a person.
37,948
the subbands. smoothing The of the of highest coefficients method is a on based
The method is based on a smoothing of the coefficients of the highest subbands.
37,949
the methods favorably most Denoising results the of use moment. compare at to in
Denoising results compare favorably to the most of methods in use at the moment.
37,950
such Gram matrix, statistics, to these second-order used textures. In describe models, as are
In these models, second-order statistics, such as Gram matrix, are used to describe textures.
37,951
further Style where to apply We create our Neural Transfer, can styles. we mixed scheme
We further apply our scheme to Neural Style Transfer, where we can create mixed styles.
37,952
experiments demonstrate results. our achieve method that The can state-of-the-art
The experiments demonstrate that our method can achieve state-of-the-art results.
37,953
robots. paper aspect explores of This visual mobile odometry the for methods deep
This paper explores the aspect of deep visual odometry methods for mobile robots.
37,954
odometry making successful. navigation this been in instrumental Visual has
Visual odometry has been instrumental in making this navigation successful.
37,955
can accurate mobile in facilitate findings. map more A robots localization for and dense
A dense map in mobile robots can facilitate for localization and more accurate findings.
37,956
proposed show benchmark datasets FGVC our efficacy the on of method. Experiments
Experiments on benchmark FGVC datasets show the efficacy of our proposed method.
37,957
is the greatly limited deep methods. metric performance of However, by sampling traditional learning
However, the performance of deep metric learning is greatly limited by traditional sampling methods.
37,958
Due the video emergence Generative exceptional witnessed Networks, has breakthroughs. to of synthesis Adversarial
Due to the emergence of Generative Adversarial Networks, video synthesis has witnessed exceptional breakthroughs.
37,959
methods proper However, in control dynamics representation videos. to lack the existing explicitly a
However, existing methods lack a proper representation to explicitly control the dynamics in videos.
37,960
standardized A individual templates. central task data to volumes to computational map is
A central computational task is to map individual data volumes to standardized templates.
37,961
the of diffeomorphic template. the subgroup acting group form stabilizer the on These
These form the stabilizer subgroup of the diffeomorphic group acting on the template.
37,962
of provides estimates form. volume the reduced-variance This
This provides reduced-variance estimates of the volume form.
37,963
derive the for variances the deformations. in bounds small We of limit Cramer-Rao
We derive Cramer-Rao bounds for the variances in the limit of small deformations.
37,964
selected each public the detailed are The for provided use. annotations of categories for
The detailed annotations for each of the selected categories are provided for public use.
37,965
Specifically, learns field the generative i.e. that propose a network we correspondence to predict deep
Specifically, we propose a deep generative network that learns to predict the correspondence field i.e.
37,966
map a image views camera between stereo in setting. disparity the calibrated two
the disparity map between two image views in a calibrated stereo camera setting.
37,967
capability long been has inferences processes desired. regarding physical explainable making of The
The capability of making explainable inferences regarding physical processes has long been desired.
37,968
fundamental motion. object One physical is process
One fundamental physical process is object motion.
37,969
However, object qualitative motion three-dimensional has been suitable theory no space. proposed there for in
However, there has been no suitable qualitative theory proposed for object motion in three-dimensional space.
37,970
to objects. what causes action infer the We of these want movement
We want to infer what action causes the movement of these objects.
37,971
multi-center studies acquisition a variety protocols. necessarily wide of initiatives are Modern with neuroimaging
Modern neuroimaging studies are necessarily multi-center initiatives with a wide variety of acquisition protocols.
37,972
paradigm the of developments restoration. learning image revolutionized have in deep Recent
Recent developments in deep learning have revolutionized the paradigm of image restoration.
37,973
fully hence differentiable end-to-end trained. proposed The is can pipeline be and restoration
The proposed restoration pipeline is fully differentiable and hence can be end-to-end trained.
37,974
given results Experimental less prone is to our data. training demonstrate work that small overfitting
Experimental results demonstrate that our work is less prone to overfitting given small training data.
37,975
is biomarker linked Cell function. that an shape important directly is to cell
Cell shape is an important biomarker that is directly linked to cell function.
37,976
temporal much less cell cell understood. morphodynamics, is namely fluctuation the shape However, of
However, cell morphodynamics, namely the temporal fluctuation of cell shape is much less understood.
37,977
cell evolution a of temporal into As mapped is result, morphodynamics morphological phenotypes.
As a result, cell morphodynamics is mapped into temporal evolution of morphological phenotypes.
37,978
Rho-signaling enhances and that the manipulating transitions. find phenotype We morphodynamics
We find that manipulating Rho-signaling enhances the morphodynamics and phenotype transitions.
37,979
vehicles. perception and control of networks have components become neural Deep key self-driving
Deep neural perception and control networks have become key components of self-driving vehicles.
37,980
consists a introspective approach explanations two parts. We of to new propose which
We propose a new approach to introspective explanations which consists of two parts.
37,981
The controller's that the output. image identifies attention regions influence network's potentially
The controller's attention identifies image regions that potentially influence the network's output.
37,982
model textual actions. of video-to-text we explanations Second, to produce use attention-based model an
Second, we use an attention-based video-to-text model to produce textual explanations of model actions.
37,983
weak-alignment. strong- attention explore alignment, two to We and approaches
We explore two approaches to attention alignment, strong- and weak-alignment.
37,984
However, and pleasing required. a explanation is visually class-discriminative more
However, a more class-discriminative and visually pleasing explanation is required.
37,985
method multi-scale incorporate difference this that prediction is model-agnostic. We a regional concept into
We incorporate this regional multi-scale concept into a prediction difference method that is model-agnostic.
37,986
regression and suppression rely Modern objects. detectors object non-maximum bounding CNN-based localize box on to
Modern CNN-based object detectors rely on bounding box regression and non-maximum suppression to localize objects.
37,987
labels reflect classification class While confidence absent. the probabilities is confidence, for localization naturally
While the probabilities for class labels naturally reflect classification confidence, localization confidence is absent.
37,988
makes during regression or suppressed bounding iterative during even NMS. boxes localized properly This degenerate
This makes properly localized bounding boxes degenerate during iterative regression or even suppressed during NMS.
37,989
A is to of a kernel used separation. between better version class MMC obtain
A kernel version of MMC is used to obtain a better between class separation.
37,990
on sidekick introduce learning of policy capitalize imbalance to We this observability.
We introduce sidekick policy learning to capitalize on this imbalance of observability.
37,991
of visualization resulting policies, interpretation a develop To technique. support the policy novel we also
To support interpretation of the resulting policies, we also develop a novel policy visualization technique.
37,992
images applications. binocular stereo a Disparity finds estimation range for of wide
Disparity estimation for binocular stereo images finds a wide range of applications.
37,993
and work supervised semantic in The both cues unsupervised manners. well
The semantic cues work well in both unsupervised and supervised manners.
37,994
scheme, data generate active In users sparsity. transmit packets the time-slotted random and RSL-MUD with
In the time-slotted RSL-MUD scheme, active users generate and transmit data packets with random sparsity.
37,995
location vision not require does and sensors than additional SafeDrive data. other
SafeDrive does not require additional sensors other than vision and location data.
37,996
non-invasive technique diseases. is used The the detection electrocardiogram a (ECG) for cardiac widely of
The electrocardiogram (ECG) is a non-invasive technique widely used for the detection of cardiac diseases.
37,997
is face This super-resolution. on paper image and
This paper is on image and face super-resolution.
37,998
mean both In squared error network's variance. has our the cases, lowest
In both cases, our network's mean squared error has the lowest variance.
37,999