Irrelevant Categories Research Articles

CLIP-SP: Vision-language model with adaptive prompting for scene parsing

We present a novel framework, CLIP-SP, and a novel adaptive prompt method to leverage pre-trained knowledge from CLIP for scene parsing. Our approach addresses the limitations of DenseCLIP, which demonstrates the superior image segmentation provided by CLIP pre-trained models over ImageNet pre-trained models, but struggles with rough pixel-text score maps for complex scene parsing. We argue that, as they contain all textual information in a dataset, the pixel-text score maps, i.e., dense prompts, are inevitably mixed with noise. To overcome this challenge, we propose a two-step method. Firstly, we extract visual and language features and perform multi-label classification to identify the most likely categories in the input images. Secondly, based on the top-k categories and confidence scores, our method generates scene tokens which can be treated as adaptive prompts for implicit modeling of scenes, and incorporates them into the visual features fed into the decoder for segmentation. Our method imposes a constraint on prompts and suppresses the probability of irrelevant categories appearing in the scene parsing results. Our method achieves competitive performance, limited by the available visual-language pre-trained models. Our CLIP-SP performs 1.14% better (in terms of mIoU) than DenseCLIP on ADE20K, using a ResNet-50 backbone.

Assessment of Transoral Robotic Surgery (TORS) Content on YouTube.

Online video hosting websites such as YouTube have been increasingly used by medical institutions to spread information about new and exciting topics. However, due to the large number of videos uploaded dailyand the lack of peer review, few attempts have been made to assess the quantity and quality of information that is uploaded on YouTube.For this study, our team assessed the available content on the transoral robotic surgery (TORS) procedure. A qualitative case study model was employed. Videos related to TORS were collected using a unified search protocol. Each video was then analyzed, and metrics of the following data points were collected: views, likes, comments, upload date, length of video, author type, author, and region of origin. Each dataset was analyzed by two distinct authors, and interrater reliability was calculated. Quantitative and qualitative statistics were curated. A total of 124 videos were analyzed for this review. The breakdown of videos was as follows: 15.32% (19) in the educational for patients category, 16.94% (21) in the educational for traineescategory, 30.65% (38) in the procedural overview category, 8.87% (11) in the patient experience (PE) category, 10.48% (13) in the promotional category, 12.10% (15) in the othercategory, and 5.65% (7) in the irrelevant (IR) category. The total number of views across all videos analyzed was 2,589,561. The total number of likes was 14,827, and the total number of comments was 2,606. The average video length was 8.63 minutes. The most viewed category was the PEcategory at 1,014,738 and the most liked at 1,714. The least viewed category was IRat 21,082. The PE category had the most engagement based on combined comments and likes. The most watched video, with 774,916 views, was in the PE category under the "TORS for Thyroidectomy" search term and was titled "Thyroid Surgery (Thyroidectomy)." As the prevalence of online videos regarding medical devices, procedures, and treatments increases, patients and trainees alike will look towardresources such as YouTube to augment their understanding. Patients, providers, and medical education platforms should take heed of the promise and pitfalls of medical content on YouTube.

Concurrent response and action effect representations across the somatomotor cortices during novel task preparation

Instructions allow us to fulfill novel and complex tasks on the first try. This skill has been linked to preparatory brain signals that encode upcoming demands in advance, facilitating novel performance. To deepen insight into these processes, we explored whether instructions pre-activated task-relevant motoric and perceptual neural states. Critically, we addressed whether these representations anticipated activity patterns guiding overt sensorimotor processing, which could reflect that internally simulating novel tasks facilitates the preparation. To do so, we collected functional magnetic resonance imaging data while female and male participants encoded and implemented novel stimulus-response associations. Participants also completed localizer tasks designed to isolate the neural representations of the mappings-relevant motor responses, perceptual consequences, and stimulus categories. Using canonical template tracking, we identified whether and where these sensorimotor representations were pre-activated. We found that response-related templates were encoded in advance in regions linked with action control, entailing not only the instructed responses but also their somatosensory consequences. This result was particularly robust in primary motor and somatosensory cortices. While, following our predictions, we found a systematic decrease in the irrelevant stimulus templates’ representational strength compared to the relevant ones, this difference was due to below-zero estimates linked to the irrelevant category activity patterns. Overall, our findings reflect that instruction processing relies on the sensorimotor cortices to anticipate motoric and kinesthetic representations of prospective action plans, suggesting the engagement of motor imagery during novel task preparation. More generally, they stress that the somatomotor system could participate with higher-level frontoparietal regions during anticipatory task control.

Public Perceptions and Discussions of the US Food and Drug Administration's JUUL Ban Policy on Twitter: Observational Study.

On June 23, 2022, the US Food and Drug Administration announced a JUUL ban policy, to ban all vaping and electronic cigarette products sold by Juul Labs. This study aims to understand public perceptions and discussions of this policy using Twitter (subsequently rebranded as X) data. Using the Twitter streaming application programming interface, 17,007 tweets potentially related to the JUUL ban policy were collected between June 22, 2022, and July 25, 2022. Based on 2600 hand-coded tweets, a deep learning model (RoBERTa) was trained to classify all tweets into propolicy, antipolicy, neutral, and irrelevant categories. A deep learning model (M3 model) was used to estimate basic demographics (such as age and gender) of Twitter users. Furthermore, major topics were identified using latent Dirichlet allocation modeling. A logistic regression model was used to examine the association of different Twitter users with their attitudes toward the policy. Among 10,480 tweets related to the JUUL ban policy, there were similar proportions of propolicy and antipolicy tweets (n=2777, 26.5% vs n=2666, 25.44%). Major propolicy topics included "JUUL causes youth addition," "market surge of JUUL," and "health effects of JUUL." In contrast, major antipolicy topics included "cigarette should be banned instead of JUUL," "against the irrational policy," and "emotional catharsis." Twitter users older than 29 years were more likely to be propolicy (have a positive attitude toward the JUUL ban policy) than those younger than 29 years. Our study showed that the public showed different responses to the JUUL ban policy, which varies depending on the demographic characteristics of Twitter users. Our findings could provide valuable information to the Food and Drug Administration for future electronic cigarette and other tobacco product regulations.

Partial Domain Adaptation Without Domain Alignment.

Unsupervised domain adaptation (UDA) aims to transfer knowledge from a well-labeled source domain to a related and unlabeled target domain with identical label space. The main workhorse in UDA is domain alignment and has proven successful. However, it is practically difficult to find an appropriate source domain with identical label space. A more practical scenario is partial domain adaptation (PDA) where the source label space subsumes the target one. Unfortunately, due to the non-identity between label spaces, it is extremely hard to obtain an ideal alignment, conversely, easier resulting in mode collapse and negative transfer. These motivate us to find a relatively simpler alternative to solve PDA. To achieve this, we first explore a theoretical analysis, which says that the target risk is bounded by both model smoothness and between-domain discrepancy. Then, we instantiate the model smoothness as an intra-domain structure preserving (IDSP) while giving up possibly riskier domain alignment. To our best knowledge, this is the first naive attempt for PDA without alignment. Finally, our empirical results on benchmarks demonstrate that IDSP is not only superior to the PDA SOTAs (e.g., ∼+10% on Cl → Rw and ∼+8% on Ar → Rw), but also complementary to domain alignment in the standard UDA.

Target Category Agnostic Knowledge Distillation With Frequency-Domain Supervision

Existing knowledge distillation approaches require task-related data to train portable student networks for satisfactory performance. Nevertheless, in real-world applications, a majority of the data is unavailable over concerns about personal privacy, commercial confidentiality, etc. To mitigate the difficulty of acquiring the target dataset, the mainstream knowledge distillation methods generate training samples from the teacher network. However, the generated images still differ from the authentic ones, which limits the student network performance. To solve this issue, we propose a convenient and cost-negligible method to build datasets for training student networks. Specifically, we crawl the data from the web without knowing any category of the target dataset, named Irrelevant Category Crawler Data (ICCD). To prevent the performance collapse due to the data distribution gap between our ICCD and the target dataset, we propose a Pseudo-Classification strategy and Frequency-domain Supervision (PCFS) for target-category agnostic knowledge distillation with ICCD. The Pseudo-Classification strategy classifies ICCD into different pseudo-categories by the teacher network, and uniformly but randomly samples images from each pseudo-category to construct the pseudo-target dataset. Furthermore, we transform the feature maps from the spatial-domain to the frequency-domain, and utilize the high-frequency and low-frequency signals of the teacher network to impose strong constraints on the student network. Extensive experiments conducted on various test sets demonstrate the effectiveness of our proposed PCFS, which outperforms existing data-free methods and achieves comparable performance to those using the target training set. Code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/SCUT-BIP-Lab/PCFS-DFKD</uri> .

EMG-FRNet: A feature reconstruction network for EMG irrelevant gesture recognition.

With the development of deep learning technology, gesture recognition based on surface electromyography (EMG) signals has shown broad application prospects in various human-computer interaction fields. Most current gesture recognition technologies can achieve high recognition accuracy on a wide range of gesture actions. However, in practical applications, gesture recognition based on surface EMG signals is susceptible to interference from irrelevant gesture movements, which affects the accuracy and security of the system. Therefore, it is crucial to design an irrelevant gesture recognition method. This paper introduces the GANomaly network from the field of image anomaly detection into surface EMG-based irrelevant gesture recognition. The network has a small feature reconstruction error for target samples and a large feature reconstruction error for irrelevant samples. By comparing the relationship between the feature reconstruction error and the predefined threshold, we can determine whether the input samples are from the target category or the irrelevant category. In order to improve the performance of EMG irrelevant gesture recognition, this paper proposes a feature reconstruction network named EMG-FRNet for EMG irrelevant gesture recognition. This network is based on GANomaly and incorporates structures such as channel cropping (CC), cross-layer encoding-decoding feature fusion (CLEDFF), and SE channel attention (SE). In this paper, Ninapro DB1, Ninapro DB5 and self-collected datasets were used to verify the performance of the proposed model. The Area Under the receiver operating characteristic Curve (AUC) values of EMG-FRNet on the above three datasets were 0.940, 0.926 and 0.962, respectively. Experimental results demonstrate that the proposed model achieves the highest accuracy among related research.

DisDSS: a novel Web-based smart disaster management system for determining the nature of a social media message for decision-making using deep learning – case study of COVID-19

Purpose This paper aims to propose DisDSS: a Web-based smart disaster management (DM) system for decision-making that will assist disaster professionals in determining the nature of disaster-related social media (SM) messages. The research classifies the tweets into need-based, availability-based, situational-based, general and irrelevant categories and visualizes them on a web interface, location-wise. Design/methodology/approach It is worth mentioning that a fusion-based deep learning (DL) model is introduced to objectively determine the nature of an SM message. The proposed model uses the convolution neural network and bidirectional long short-term memory network layers. Findings The developed system leads to a better performance in accuracy, precision, recall, F-score, area under receiver operating characteristic curve and area under precision-recall curve, compared to other state-of-the-art methods in the literature. The contribution of this paper is three fold. First, it presents a new covid data set of SM messages with the label of nature of the message. Second, it offers a fusion-based DL model to classify SM data. Third, it presents a Web-based interface to visualize the structured information. Originality/value The architecture of DisDSS is analyzed based on the practical case study, i.e. COVID-19. The proposed DL-based model is embedded into a Web-based interface for decision support. To the best of the authors’ knowledge, this is India’s first SM-based DM system.

A bibliometric analysis deconstructing research on how digitalisation affects knowledge workers

Whereas previous waves of automation have predominantly affected labour intensive work, current cognitive technologies enable automation and commoditization of knowledge intensive work. However, extant research provides limited understanding of how this man-machine collaboration will unfold. Moreover, knowledge intensive firms, such as professional service firms, are characterized by how professionals show a preference for autonomy, exhibiting a distaste for control, supervision, and formal organizational processes. Therefore, it is particularly prudent to investigate how knowledge workers will be affected by the introduction of technologies like AI, IoT and Big Data, giving special attention to their professional autonomy. However, there exist limited empirical studies on how the interaction between autonomous professionals and machines unfold in practice, and there exist no unifying theory explaining the effect of digitalization on this type of knowledge intensive work. To provide a starting point for subsequent research, we offer a structured literature review aided by a bibliometric analysis to identify core contributions that can identify and synthesize main dimensions in extant research. We experimented with different searches phrases at Web of Science, and ended up with combining (Digi* OR techno*) AND autonomy, which resulted in 3961 articles. By excluding irrelevant categories our result was 815 articles. We then performed a bibliometric analysis aided by VOSviewer and narrowed our dataset down to 15 core articles. A content analysis of these core articles identified three different goals of digital implementation; digital tools to enhance communication and sharing of information, digital tools for better decisions and digital tools for monitoring and control. We found paradoxical relations regarding worker autonomy and technology within all three goals. The content analysis points in the direction that the increased control following the new technology dominated the gained autonomy for knowledge professionals. We found three dominating factors that could be used to explain how the professionals perceived the change in autonomy. These are; what kind of job, what kind of technology and the professional’s attitude towards social status, technology, change and knowledge sharing. We suggest that practitioners who are implementing digital technologies must pay attention to relations that affect knowledge professionals perceived autonomy.

Influence of target-distractor neural similarity on working memory performance in older and younger adults

ABSTRACT According to the inhibitory deficit hypothesis, older adults often fail to selectively inhibit distractors and attend to relevant information in working memory, leading to poorer memory of target items but better recall of irrelevant distractors compared to younger adults. Here, we explored how neural similarity of activity patterns between relevant and irrelevant stimulus categories impacts memory performance. We found evidence that older adults may benefit from failing to inhibit distractors that are similar to targets, perhaps because sustained neural activation of distractors partially supports maintenance of targets when they share neural resources, allowing for better subsequent recognition of studied target items. We also found increased category-specific multivoxel pattern activity in medial temporal regions in younger compared to older adults as category similarity increased. We propose that this reduced category-specific activation in medial temporal regions in older adults may reflect more blended representations of all the information available in working memory.

Finding Nonrigid Tiny Person With Densely Cropped and Local Attention Object Detector Networks in Low-Altitude Aerial Images

Finding tiny persons under the drone vision was, is and remains to be an integral and challenging task. Unmanned Aerial Vehicles (UAVs) with high-speed, low-altitude, and multi-perspective flight bring about violently various scales of objects, which burdens the optimization of models. Moreover, the detection performance of densely and faintly discernible person characteristics is far less than that of large objects in high-resolution aerial images. In this paper, we introduce the image cropping strategy and attention mechanism based on YOLOv5 to address small person detection in the optimized VisDrone2019 dataset. Specifically, we propose a Densely Cropped and Local Attention of object detector Network (DCLANet), which is inspired by the observation that less area occupied by small objects should be fully focused and relatively magnified in the original image. DCLANet assembled Density Map Guided Object Detection (DMNet) in Aerial Images and You Only Look Twice (YOLT): Rapid Multi-Scale Object Detection In Satellite Imagery to crop images upon training and testing stage, meanwhile, added Bottleneck Attention Mechanism (BAM) to YOLOv5 baseline framework, which more focus on person objects other than irrelevant categories. To achieve further improvement of DCLANet, we also provide bags of useful strategies: data augmentation, label fusion, category filtering and hyperparameter evolution. Extensive experiments on the VisDrone2019 show that DCLANet achieves state-of-the-art performance, the detection result of person category <formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex>$A P^{\rm{val}}$</tex></formula> @0.5 is 50.04% with test-dev subset, which is substantially better than the previous SOTA method(DPNetV3) by 12.01%. In addition, on our optimized VisDrone2019 dataset, <formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex>$A P^{\rm{val}}$</tex></formula> @0.5 and A P test@0.5 obtained 74.95%, 62.18%, respectively. Compared to YOLOv5, DCLANet improves 3.8% or so, which is encouraging and competitve.

Innovative ratiometric optical strategy: Nonconjugated polymer dots based fluorescence-scattering dual signal output for sensing mercury ions

A new ratiometric platform was developed for sensing Hg2+, which combined fluorescence and scattering simultaneously. This ratiometric strategy reflected superiorities over conventional methods, since the two independent signals at irrelevant categories meet the requirements of sufficient wavelength separation, stimulation under one excitation, and collection on single instrument. Herein, nonconjugated polymer dots (N-PDs) were served as the recognition unit for Hg2+ with turn-off fluorescence and turn-on scattering. Additionally, two signal collection tactics were proposed to achieve fluorescence and scattering in a window: one was to record down-conversion fluorescence and second-order scattering spectra (FL@SOS), and the other was to gather the fluorescence excited by second-order diffraction light and first-order scattering (SODL-FL@FOS). This ratiometric sensor exhibited outstanding performance toward Hg2+ in the range of 0.1–50 μM with the detection limit of 27 nM. By contrast, the present proposal provided a more ingenious and scalable way to construct ratiometric sensor than traditional approach.

Odds favor the bees.

Muszynski and Couvillon (see record 2020-37265-001) built upon their previous findings that honeybees can learn the relation among triads of trial-unique visual stimuli. In this new work, they showed that bees encountering trial-unique sets of three or four visual stimuli chose the correct stimulus at above-chance levels, replicating their previous findings and extending them to four-choice displays. In the first experiment, the bees' performance with triads of stimuli was unaffected by whether the correct choice was patterned or solid, or whether the stimuli shared a common color. A control group in this experiment encountered a categorical discrimination problem with two stimuli. This latter group of bees easily learned the discrimination and made a lower proportion of errors than bees solving the oddity problem, suggesting that the bees did not perceive the oddity task as a discrimination problem. The possibility that bees solved the oddity problem as a categorical discrimination was further examined in a second experiment. In that experiment, one group of bees encountered quartets of disks in combinations of solid color and two-color disks, and another group encountered only two-color disks. The authors expected that the addition of an irrelevant category (solid or two-color disk) would make the odd stimulus more discriminable, and, therefore, improve performance in that group compared with the group that encountered only two-colored disks. Their expectation was confirmed: Bees that encountered stimuli with a categorical difference, even though the category was irrelevant to which disk (of four) was odd, averaged more correct choices (average .67 vs. .47 across 15 trials; .25 expected by chance) and reached a higher terminal level of performance than bees that encountered only two-color disks (nearing .90 vs. around .50 correct, Trials 14 -16, solid and pattern group vs. pattern-only group, respectively). (PsycInfo Database Record (c) 2020 APA, all rights reserved).

Category difference facilitates oddity learning in honeybees (Apis mellifera).

Performance of honeybees resembles that of vertebrates in a variety of associative learning experiments. Recent work has focused on relational learning phenomena not easily explained by associative principles, including same/different problems, the simplest of which is the oddity problem. Free-flying bees were trained to visit a laboratory window and were rewarded for choice of the odd stimulus among a set of stimuli. There were two stimulus categories, single-color solids and two-color patterns. The training was trial-unique, with new sets of stimuli on each trial. In Experiment 1, 4 groups were trained in a 3-stimulus oddity problem, 2 with solid odd and patterns nonodd and 2 with pattern odd and solids nonodd. For 1 group in each condition, the odd and nonodd stimuli shared a color. The performance of all groups was better than chance. The bees could solve the problem on the basis of oddity (same vs. different) or category (solid vs. pattern). These possibilities were unconfounded in Experiment 2 with 2 groups trained in a 4-stimulus oddity problem. Group 1 was trained with a category difference on each trial; the solid color was odd on half the trials and the pattern odd on the others. Group 2 was trained with no category difference; all stimuli were patterns. Both groups showed better-than-chance performance, and the irrelevant category difference facilitated oddity discrimination for Group 1. The results support previous findings of oddity learning in honeybees, the only invertebrate species for which any relational learning phenomena have been demonstrated. (PsycInfo Database Record (c) 2020 APA, all rights reserved).

Exploration of the correlation between GPCRs and drugs based on a learning to rank algorithm

Exploring the protein - drug correlation can not only solve the problem of selecting candidate compounds but also solve related problems such as drug redirection and finding potential drug targets. Therefore, many researchers have proposed different machine learning methods for prediction of protein-drug correlations. However, many existing models simply divide the protein-drug relationship into related or irrelevant categories and do not deeply explore the most relevant target (or drug) for a given drug (or target). In order to solve this problem, this paper applies the ranking concept to the prediction of the GPCR (G Protein-Coupled Receptors)-drug correlation. This study uses two different types of data sets to explore candidate compound and potential target problems, and both sets achieved good results. In addition, this study also found that the family to which a protein belongs is not an inherent factor that affects the ranking of GPCR-drug correlations; however, if the drug affects other family members of the protein, then the protein is likely to be a potential target of the drug. This study showed that the learning to rank algorithm is a good tool for exploring protein-drug correlations.

P300 and positive slow waves reveal the plausibility in inductive reasoning.

Category-based induction is an advanced cognitive function that is based on our category-level knowledge. Previous findings have recognized the distance effect in category-based induction: Inductive strength is affected by the hierarchical distance between the premises and conclusions. However, the neural mechanisms underlying this effect require elucidation. In the present study, we investigated the neural mechanisms of the distance effect by using EEG technology and a new experimental paradigm-category-based induction. In this paradigm, we used three hierarchical levels of categories-the subordinate category, the basic category, the superordinate category-and an irrelevant category. We further used these categories to create four types of trial that varied in the hierarchical distance between the premise and the conclusion: the subordinate-basic, the basic-superordinate, the subordinate-superordinate, and the irrelevant-superordinate trials. In each trial, participants judged the probability that the conclusion category had the same property as the premise category. Our behavioral results revealed that people responded more slowly in the irrelevant-superordinate trials than in the basic-superordinate and the subordinate-basic trials. Our ERP results showed that the irrelevant-superordinate trials elicited smaller P300 (250-500ms) amplitudes than did the subordinate-basic and the basic-superordinate trials. In addition, the subordinate-superordinate trials elicited smaller P300 and PSW (700-998ms) amplitudes than did the subordinate-basic and the basic-superordinate combinations. These findings indicate that the amplitudes of P300 and PSW may reflect the distance effect in inductive reasoning: The further the premise-conclusion hierarchical distance, the lower the inductive strength, and thus the smaller the P300 and PSW amplitudes.

Predicting primary categories of business listings for local search ranking

We consider the problem of identifying primary categories of a business listing among the categories provided by the owner of the business, in order to enhance local search and browsing. The category information submitted by business owners cannot be trusted with absolute certainty since they may purposefully add some secondary or irrelevant categories to increase recall in local search results, which makes category search very challenging for local search engines. Thus, identifying primary categories of a business is a crucial problem in local search. This problem can be cast as a multi-label classification problem with a large number of categories. However, the large scale of the problem makes it infeasible to use conventional supervised-learning-based text categorization approaches.We propose a large-scale classification framework that leverages multiple types of classification labels to produce a highly accurate classifier with fast training time. We effectively combine the complementary label sources to refine prediction. The experimental results indicate that our framework achieves very high precision and recall and outperforms a competitive baseline using a centroid-based method.We also propose a new ranking feature based on the mapping of queries and documents to category space and show that the new feature leads to ranking relevance improvements for local search.

Intentional Memory Instructions Direct Attention But Do Not Enhance Visual Memory

1. D. Alexander Varakin[1][1] 2. Jamie Hale[1][1] 1. 1Eastern Kentucky University, Richmond, USA 1. D. Alexander Varakin, Department of Psychology, Eastern Kentucky University, 127 Cammack Building, 521 Lancaster Ave., Richmond, KY 40475, USA. Email: donald.varakin{at}eku.edu The current experiment investigated the role of intentional-encoding instructions and task relevance at study on visual memory performance. Task relevance was manipulated by having participants keep a running tally of either the to-be-remembered objects or an irrelevant category of objects during study. Half of the participants within each level of task relevance were further instructed to remember one category of objects for a subsequent recognition test, and the other half of the participants were not informed of a memory test. Intentional-encoding instructions improved recognition discrimination only when participants were not already keeping a tally of the to-be-remembered objects. This result suggests that intentional-encoding instructions may improve visual memory via generic attentional modulation, not encoding-specific processes. [1]: #aff-1

Content based Image Retrieval System with Watermarks and Relevance Feedback

Image retrieval and related operations are always a ’hotspot’ in the information era.Content-based image retrieval (CBIR) is a vastly developing area in the multimedia technology domain. To enhance security, we apply watermarking technique into the retrieval system and propose an approach for JEPG image retrieval. The proposed image retrieval system consists of three main phases, offline process, online retrieval process and the feedback process. The offline process aims at the feature vector extraction from the image. Later these features will be stored in the database. When it comes to the online retrieval process, it actually extracts the image features from the input image and matches these feature vectors with those available in the image database. In order to overcome the possible dissimilarity between bottom features and high-level semantics in the image retrieval; we introduce the feedback network to strengthen the retrieval efficiency.This is a simple categorized screening.Such a feedback scenario makes the system more user-friendly and effective.The proposed feedback screening strategy filters the images from irrelevant categories and enriches the final result with more relevant images.

Cascade category-aware visual search.

Incorporating image classification into image retrieval system brings many attractive advantages. For instance, the search space can be narrowed down by rejecting images in irrelevant categories of the query. The retrieved images can be more consistent in semantics by indexing and returning images in the relevant categories together. However, due to their different goals on recognition accuracy and retrieval scalability, it is hard to efficiently incorporate most image classification works into large-scale image search. To study this problem, we propose cascade category-aware visual search, which utilizes weak category clue to achieve better retrieval accuracy, efficiency, and memory consumption. To capture the category and visual clues of an image, we first learn category-visual words, which are discriminative and repeatable local features labeled with categories. By identifying category-visual words in database images, we are able to discard noisy local features and extract image visual and category clues, which are hence recorded in a hierarchical index structure. Our retrieval system narrows down the search space by: 1) filtering the noisy local features in query; 2) rejecting irrelevant categories in database; and 3) preforming discriminative visual search in relevant categories. The proposed algorithm is tested on object search, landmark search, and large-scale similar image search on the large-scale LSVRC10 data set. Although the category clue introduced is weak, our algorithm still shows substantial advantages in retrieval accuracy, efficiency, and memory consumption than the state-of-the-art.