Inflammatory factor-based prognostic risk stratification for patients with metastatic castration-resistant prostate cancer treated with docetaxel.

To develop a simple inflammatory factor-based prognostic risk stratification system for patients with metastatic castration-resistant prostate cancer (mCRPC) receiving docetaxel as the initial treatment, we reviewed the data of 399 consecutive patients who received first-line docetaxel chemotherapy between January 2013 and June 2019 retrospectively. The optimal cut-off values for the neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR) in terms of survival were calculated by ROC curves. Patients were stratified into favourable (lower NLR and lower PLR), intermediate (higher NLR and lower PLR, or lower NLR and higher PLR) and poor (higher NLR and higher PLR) groups. Kaplan-Meier curves were drawn to evaluate overall survival (OS) and progression-free survival (PFS). The ROC curve analysis determined the cut-offs for the NLR and PLR to be 2.355 and 104.275 respectively. Multivariate Cox regression analysis showed that being in the poor patient group (NLR ≥2.355 and PLR ≥104.275) was an independent prognostic risk factor and Kaplan-Meier curves analysis revealed that respondents with NLR <2.355 and PLR <104.275 had significantly longer OS and PFS. So it can be concluded that concurrently high NLR and PLR values are predictors for poor chemotherapy outcomes after androgen deprivation therapy failure in patients with mCRPC.

Tetrahydroxystilbene glucoside relieves the chronic inflammatory pain by inhibiting neuronal apoptosis, microglia activation, and GluN2B overexpression in anterior cingulate cortex.

Tetrahydroxystilbene glucoside (THSG) is one of the active ingredients of Polygonum multiflorum. It has been shown to exert a variety of pharmacological effects, including antioxidant, anti-aging, and anti-atherosclerosis. Because of its prominent anti-inflammatory effect, we explored whether THSG had analgesic effect. In this study, we used a model of chronic inflammatory pain caused by injecting complete Freund's adjuvant into the hind paw of mice. We found THSG relieved swelling and pain in the hind paw of mice on a dose-dependent manner. In the anterior cingulate cortex, THSG suppressed the upregulation of GluN2B-containing N-methyl-D-aspartate receptors and the downregulation of GluN2A-containing N-methyl-D-aspartate receptors caused by chronic inflammation. In addition, THSG increased Bcl-2 and decreased Bax and Caspase-3 expression by protecting neuronal survival. Furthermore, THSG inhibited the phosphorylation of p38 and the increase of nuclear factor κB (NF-κB) and tumor necrosis factor α (TNF-α). Immunohistochemical staining revealed that THSG blocked the activation of microglia and reduced the release of proinflammatory cytokines TNF-α, interleukin 1ß (IL-1ß), and interleukin 6 (IL-6). In conclusion, this study demonstrated that THSG had a certain effect on alleviating complete Freund's adjuvant-induced chronic inflammatory pain.

LAG-3 Represents a Marker of CD4+ T Cells with Regulatory Activity in Patients with Bone Fracture.

The lymphocyte activation gene 3 (LAG-3) is a CD4 homolog with binding affinity to MHC class II molecules. It is thought that LAG-3 exerts a bimodal function, such that co-ligation of LAG-3 and CD3 could deliver an inhibitory signal in conventional T cells, whereas, on regulatory T cells, LAG-3 expression could promote their inhibitory function. In this study, we investigated the role of LAG-3 expression on CD4+ T cells in patients with long bone fracture. We found that LAG-3+ cells represented approximately 13% of peripheral blood CD4+ T cells on average. Compared to LAG-3- CD4+ T cells, LAG-3+ CD4+ T cells presented significantly higher Foxp3 and CTLA-4 expression. Directly ex vivo or with TCR stimulation, LAG-3+ CD4+ T cells expressed significantly higher levels of IL-10 and TGF-ß than LAG-3- CD4+ T cells. Interestingly, blocking the LAG-3-MHC class II interaction actually increased the IL-10 expression by LAG-3+ CD4+ T cells. The frequency of LAG-3+ CD4+ T cell was positively correlated with restoration of healthy bone function in long bone fracture patients. These results together suggested that LAG-3 is a marker of CD4+ T cells with regulatory function; at the same time, LAG-3 might have limited the full suppressive potential of Treg cells.

FK506 Attenuates the Inflammation in Rat Spinal Cord Injury by Inhibiting the Activation of NF-κB in Microglia Cells.

FK-506 (Tacrolimus) is a very commonly used immunomodulatory agent that plays important roles in modulating the calcium-dependent phosphoserine-phosphothreonine protein phosphatase calcineurin and thus inhibits calcineurin-mediated secondary neuronal damage. The biological function of FK-506 in the spinal cord has not been fully elucidated. To clarify the anti-inflammatory action of FK-506 in spinal cord injury (SCI), we performed an acute spinal cord contusion injury model in adult rats and hypoxia-treated primary spinal cord microglia cultures. This work studied the activation of NF-κB and proinflammatory cytokine (TNF-a, IL-1b, and IL-6) expression. ELISA and q-PCR analysis revealed that TNF-a, IL-1b, and IL-6 levels significantly increased 3 days after spinal cord contusion and decreased after 14 days, accompanied by the increased activation of NF-κB. This increase was reversed by an FK-506 treatment. Double immunofluorescence labeling suggested that NF-κB activation was especially prominent in microglia. Immunohistochemistry confirmed no alteration in the number of microglia. Moreover, the results in hypoxia-treated primary spinal cord microglia confirmed the effect of FK-506 on TNF-a, IL-1b, and IL-6 expression and NF-κB activation. These findings suggest that FK-506 may be involved in microglial activation after SCI.

On Applicability of Tunable Filter Bank Based Feature for Ear Biometrics: A Study from Constrained to Unconstrained.

In this paper, an overall framework has been presented for person verification using ear biometric which uses tunable filter bank as local feature extractor. The tunable filter bank, based on a half-band polynomial of 14th order, extracts distinct features from ear images maintaining its frequency selectivity property. To advocate the applicability of tunable filter bank on ear biometrics, recognition test has been performed on available constrained databases like AMI, WPUT, IITD and unconstrained database like UERC. Experiments have been conducted applying tunable filter based feature extractor on subparts of the ear. Empirical experiments have been conducted with four and six subdivisions of the ear image. Analyzing the experimental results, it has been found that tunable filter moderately succeeds to distinguish ear features at par with the state-of-the-art features used for ear recognition. Accuracies of 70.58%, 67.01%, 81.98%, and 57.75% have been achieved on AMI, WPUT, IITD, and UERC databases through considering Canberra Distance as underlying measure of separation. The performances indicate that tunable filter is a candidate for recognizing human from ear images.

Gender recognition from unconstrained and articulated human body.

Gender recognition has many useful applications, ranging from business intelligence to image search and social activity analysis. Traditional research on gender recognition focuses on face images in a constrained environment. This paper proposes a method for gender recognition in articulated human body images acquired from an unconstrained environment in the real world. A systematic study of some critical issues in body-based gender recognition, such as which body parts are informative, how many body parts are needed to combine together, and what representations are good for articulated body-based gender recognition, is also presented. This paper also pursues data fusion schemes and efficient feature dimensionality reduction based on the partial least squares estimation. Extensive experiments are performed on two unconstrained databases which have not been explored before for gender recognition.

Impact of Alcian blue and periodic acid Schiff expression on the prognosis of gastric signet ring cell carcinoma.

BACKGROUND: The Alcian blue (AB) and periodic acid Schiff (PAS) stains are representative mucus markers in gastric signet ring cell carcinoma (SRCC). They are low-cost special staining methods used to detect acidic mucus and neutral mucus, respectively. However, the clinical importance of the special combined AB and PAS stain is unclear. AIM: To investigate AB expression, PAS expression and the AB-to-PAS (A/P) ratio in gastric SRCC patients and to assess patient prognosis. METHODS: Paraffin-embedded sections from 83 patients with gastric SRCC were stained with AB and PAS, and signet ring cell positivity was assessed quantitatively. Immunohistochemical staining for Ki67, protein 53 (P53) and human epidermal growth factor receptor 2 (HER2) was performed simultaneously. The cancer-specific survival (CSS) rate was estimated via Kaplan-Meier analysis. Cox proportional hazards models were used for univariate and multivariate survival analyses. RESULTS: Kaplan-Meier survival analysis revealed that the 3-year CSS rate was significantly greater in the high-PAS-expression subgroup than in the low-PAS-expression subgroup (P < 0.001). The 3-year CSS rate in the A/P ≤ 0.5 group was significantly greater than that in the A/P > 0.5 group (P = 0.042). Univariate Cox regression analysis revealed that the factors affecting prognosis included tumor diameter, lymph node metastasis, vessel carcinoma embolus, tumor stage, the A/P ratio and the expression of Ki67, P53 and the PAS. Cox multivariate regression analysis confirmed that low PAS expression [hazard ratio (HR) = 3.809, 95% confidence interval (CI): 1.563-9.283, P = 0.003] and large tumor diameter (HR = 2.761, 95%CI: 1.086-7.020, P = 0.033) were independent risk factors for poor prognosis. CONCLUSION: A/P > 0.5 is potentially a risk factor for prognosis, and low PAS expression is an independent risk factor in the prognosis of gastric SRCC. PAS expression and the A/P ratio could help in predicting the clinical prognosis of patients with SRCC.

Precise lymph node biopsy for endometrial cancer confined to the uterus: Analysis of 43 clinical cases.

OBJECTIVE: To explore a precise association between tumor location and lymph node (LN) biopsy algorithm in uterine confined endometrial cancer (EC). MATERIALS AND METHODS: Patients with EC treated in the Department of Obstetrics and Gynecology, South Branch of Fujian Provincial Hospital were included in this observational retrospective study. Based on the procedure of treatment, patients were separated to stage I (2015.07-2019.09) and stage II (2019.09-2021.9). In each stage, patients were separated to high and low-risk group by the predicted results. Patients in the high-risk group received systematic lymphadenectomy in stage I and sentinel lymph node (SLN) dissection in stage II. The efficiency of lymph node metastasis (LNM) detection rates was compared between stage I and stage II cases. Precise lymph node biopsy algorithm was also constructed based on the outcomes of stage II. RESULTS: Overall, 43 patients, 28 in stage I and 15 in stage II, were included in the study. No recurrence or death cases had been found within follow-up terms. Based on the difference in the detection efficiency of LNM (p > 0.05), there was no difference between two stages. Thus, systematic lymphadenectomy and SLN biopsy provided similar success rates. The location of tumor site was also important for deciding whether pelvic or para-aortic SLN should be sampled for LNM. CONCLUSIONS: Precise SLN biopsy for EC confined to the uterus showed comparable LNM detection rate as systematic lymphadenectomy. EC location may be used to determine whether pelvic or para-aortic SLN sampling should be conducted for treatment.

CaMKII activates ASK1 to induce apoptosis of spinal astrocytes under oxygen-glucose deprivation.

Calcium/calmodulin-dependent protein kinase II (CaMKII) is a ubiquitous, structurally complex multifunctional protein serine/threonine kinase that plays an important role in cell apoptosis via linking the ER stress and mitochondrial apoptosis pathways. Recently, CaMKII has been correlated with apoptosis signal-regulating kinase 1 (ASK1) activity and the ASK1-dependent apoptosis pathway through the direct phosphorylation of Thr845 of ASK1. The specific role of CaMKII in hypoxia-reoxygenation (H/R)-induced spinal astrocyte apoptosis, however, remains unclear. In this study, we investigated the effects of CaMKIIγ (an isoform of CaMKII) on spinal astrocyte apoptosis using an in vitro oxygen-glucose deprivation (OGD/R) model which mimics hypoxic/ischemic conditions in vivo. OGD/R increased cell death and the activation of CaMKII. Deletion of CaMKIIγ results in the reduced activation of CaMKII and apoptosis in astrocytes under OGD/R conditions. Notably, the deletion of CaMKIIγ induced ASK1 phosphorylation at Thr845 in astrocytes. The activation of JNK and p38 and the downstream effect of ASK1 were also reduced. These data suggest that CaMKIIγ is required for the CaMKII-dependent regulation of ASK1, affecting the apoptosis of a biologically important cell type under spinal cord injury.

CHD1 deletion stabilizes HIF1α to promote angiogenesis and glycolysis in prostate cancer.

Chromodomain-helicase-DNA-binding protein 1 (CHD1) deletion is among the most common mutations in prostate cancer (PCa), but its role remains unclear. In this study, RNA sequencing was conducted in PCa cells after clustered regularly interspaced palindromic repeat (CRISPR)/CRISPR-associated protein 9 (Cas9)-based CHD1 knockout. Gene set enrichment analysis (GSEA) indicated upregulation of hypoxia-related pathways. A subsequent study confirmed that CHD1 deletion significantly upregulated hypoxia-inducible factor 1α (HIF1α) expression. Mechanistic investigation revealed that CHD1 deletion upregulated HIF1α by transcriptionally downregulating prolyl hydroxylase domain protein 2 (PHD2), a prolyl hydroxylase catalyzing the hydroxylation of HIF1α and thus promoting its degradation by the E3 ligase von Hippel-Lindau tumor suppressor (VHL). Functional analysis showed that CHD1 deletion promoted angiogenesis and glycolysis, possibly through HIF1α target genes. Taken together, these findings indicate that CHD1 deletion enhances HIF1α expression through PHD2 downregulation and therefore promotes angiogenesis and metabolic reprogramming in PCa.

Pneumatically Controlled Reconfigurable Bistable Bionic Flower for Robotic Gripper.

Beyond their colorful appearances and versatile geometries, flowers can self-shape-morph by adapting to environmental changes. Nature-inspired artificial systems that mimic their natural counterparts in function, flexibility, and adaptation find an emerging application in mobile robotics. In this study, a novel reconfigurable bionic flower made of petal-shaped bistable carbon fiber-reinforced composites and actuated by soft pneumatic actuators is presented. A robotic gripper based on the bionic flower was then developed for transportation tasks. First, a bionic petal based on a hybridization of bistable composites was designed and a theoretical model was established to analyze its bistable characteristic. Second, experiments and simulations were performed to analyze the out-of-plane deformation and morphing processes of the bionic petal. Curvature analysis of the closing state and blooming state shows a good match with the theoretical results. Finally, a flower-inspired robotic gripper made of the bionic petal is demonstrated to evaluate its gripping performances, including gripping force, response time, and reliability. The functional tests confirmed that the proposed soft gripper can grip objects of various shapes, sizes, and weights within milliseconds response time. The stable gripping configuration was maintained through the bistability of the bionic petal without continuous pressure consumption. The high reliability of the gripper is very useful for gripping tasks under unstructured environments, where precise control over the robot is not possible.

Cross-Batch Hard Example Mining With Pseudo Large Batch for ID vs. Spot Face Recognition.

In our daily life, a large number of activities require identity verification, e.g., ePassport gates. Most of those verification systems recognize who you are by matching the ID document photo (ID face) to your live face image (spot face). The ID vs. Spot (IvS) face recognition is different from general face recognition where each dataset usually contains a small number of subjects and sufficient images for each subject. In IvS face recognition, the datasets usually contain massive class numbers (million or more) while each class only has two image samples (one ID face and one spot face), which makes it very challenging to train an effective model (e.g., excessive demand on GPU memory if conducting the classification on such massive classes, hardly capture the effective features for bisample data of each identity, etc.). To avoid the excessive demand on GPU memory, a two-stage training method is developed, where we first train the model on the dataset in general face recognition (e.g., MS-Celeb-1M) and then employ the metric learning losses (e.g., triplet and quadruplet losses) to learn the features on IvS data with million classes. To extract more effective features for IvS face recognition, we propose two novel algorithms to enhance the network by selecting harder samples for training. Firstly, a Cross-Batch Hard Example Mining (CB-HEM) is proposed to select the hard triplets from not only the current mini-batch but also past dozens of mini-batches (for convenience, we use batch to denote a mini-batch in the following), which can significantly expand the space of sample selection. Secondly, a Pseudo Large Batch (PLB) is proposed to virtually increase the batch size with a fixed GPU memory. The proposed PLB and CB-HEM can be employed simultaneously to train the network, which dramatically expands the selecting space by hundreds of times, where the very hard sample pairs especially the hard negative pairs can be selected for training to enhance the discriminative capability. Extensive comparative evaluations conducted on multiple IvS benchmarks demonstrate the effectiveness of the proposed method.

UBE2T/STAT3 Signaling Promotes the Proliferation and Tumorigenesis in Retinoblastoma.

Purpose: The purpose of this paper was to investigate the expression and function of Ubiquitin-conjugating enzyme 2T (UBE2T), a human E2 ubiquitin-conjugating enzyme, in human retinoblastoma. Methods: The expression of UBE2T in normal retina and retinoblastoma was analyzed using the Gene Expression Omnibus (GEO) databases, and its expression was immunohistochemically evaluated in 29 retinoblastoma sections and 5 normal retinas. Then CCK-8, flow cytometry, RNA-sequencing analysis, and in vivo assays were performed to explore the exact role of UBE2T in retinoblastoma. Results: We found that retinoblastoma showed higher UBE2T expression than normal retina in GEO datasets and tissues. The immunoreactive score of UBE2T ≥4 was associated with group E in IIRC, T2-T4b in pTNM staging, poorly differentiated retinoblastoma, and high-risk histopathological factors. Knockdown of UBE2T reduced the cell viability, increased the apoptosis cells and G0/G1 cells, and inhibited subcutaneous tumor growth in vivo. Mechanistic studies showed that UBE2T knockdown induced down-regulation of phosphorylation of STAT3 and its downstream genes in vitro and in vivo. Rescue assays confirmed that STAT3 signaling pathway was involved in the effect of reduced cell viability, elevated apoptosis cells, and G0/G1 cells mediated by UBE2T knockdown. Conclusions: Our data indicate that UBE2T significantly participates in the proliferation of retinoblastoma via the STAT3 signaling pathway, suggesting the potential of UBE2T as a therapeutic target for retinoblastoma treatment.

ChaLearn Looking at People: IsoGD and ConGD Large-Scale RGB-D Gesture Recognition.

The ChaLearn large-scale gesture recognition challenge has run twice in two workshops in conjunction with the International Conference on Pattern Recognition (ICPR) 2016 and International Conference on Computer Vision (ICCV) 2017, attracting more than 200 teams around the world. This challenge has two tracks, focusing on isolated and continuous gesture recognition, respectively. It describes the creation of both benchmark datasets and analyzes the advances in large-scale gesture recognition based on these two datasets. In this article, we discuss the challenges of collecting large-scale ground-truth annotations of gesture recognition and provide a detailed analysis of the current methods for large-scale isolated and continuous gesture recognition. In addition to the recognition rate and mean Jaccard index (MJI) as evaluation metrics used in previous challenges, we introduce the corrected segmentation rate (CSR) metric to evaluate the performance of temporal segmentation for continuous gesture recognition. Furthermore, we propose a bidirectional long short-term memory (Bi-LSTM) method, determining video division points based on skeleton points. Experiments show that the proposed Bi-LSTM outperforms state-of-the-art methods with an absolute improvement of 8.1% (from 0.8917 to 0.9639) of CSR.

SMGEA: A New Ensemble Adversarial Attack Powered by Long-Term Gradient Memories.

Deep neural networks are vulnerable to adversarial attacks. More importantly, some adversarial examples crafted against an ensemble of source models transfer to other target models and, thus, pose a security threat to black-box applications (when attackers have no access to the target models). Current transfer-based ensemble attacks, however, only consider a limited number of source models to craft an adversarial example and, thus, obtain poor transferability. Besides, recent query-based black-box attacks, which require numerous queries to the target model, not only come under suspicion by the target model but also cause expensive query cost. In this article, we propose a novel transfer-based black-box attack, dubbed serial-minigroup-ensemble-attack (SMGEA). Concretely, SMGEA first divides a large number of pretrained white-box source models into several "minigroups." For each minigroup, we design three new ensemble strategies to improve the intragroup transferability. Moreover, we propose a new algorithm that recursively accumulates the "long-term" gradient memories of the previous minigroup to the subsequent minigroup. This way, the learned adversarial information can be preserved, and the intergroup transferability can be improved. Experiments indicate that SMGEA not only achieves state-of-the-art black-box attack ability over several data sets but also deceives two online black-box saliency prediction systems in real world, i.e., DeepGaze-II (https://deepgaze.bethgelab.org/) and SALICON (http://salicon.net/demo/). Finally, we contribute a new code repository to promote research on adversarial attack and defense over ubiquitous pixel-to-pixel computer vision tasks. We share our code together with the pretrained substitute model zoo at https://github.com/CZHQuality/AAA-Pix2pix.

Identification of UBE2I as a Novel Biomarker in ccRCC Based on a Large-Scale CRISPR-Cas9 Screening Database and Immunohistochemistry.

Background: The genome-wide CRISPR-cas9 dropout screening has emerged as an outstanding approach for characterization of driver genes of tumor growth. The present study aims to investigate core genes related to clear cell renal cell carcinoma (ccRCC) cell viability by analyzing the CRISPR-cas9 screening database DepMap, which may provide a novel target in ccRCC therapy. Methods: Candidate genes related to ccRCC cell viability by CRISPR-cas9 screening from DepMap and genes differentially expressed between ccRCC tissues and normal tissues from TCGA were overlapped. Weighted gene coexpression network analysis, pathway enrichment analysis, and protein-protein interaction network analysis were applied for the overlapped genes. The least absolute shrinkage and selection operator (LASSO) regression was used to construct a signature to predict the overall survival (OS) of ccRCC patients and validated in the International Cancer Genome Consortium (ICGC) and E-MTAB-1980 database. Core protein expression was determined using immunohistochemistry in 40 cases of ccRCC patients. Results: A total of 485 essential genes in the DepMap database were identified and overlapped with differentially expressed genes in the TCGA database, which were enriched in the cell cycle pathway. A total of four genes, including UBE2I, NCAPG, NUP93, and TOP2A, were included in the gene signature based on LASSO regression. The high-risk score of ccRCC patients showed worse OS compared with these low-risk patients in the ICGC and E-MTAB-1980 validation cohort. UBE2I was screened out as a key gene. The immunohistochemistry indicated UBE2I protein was highly expressed in ccRCC tissues, and a high-level nuclear translocation of UBE2I occurs in ccRCC. Based on the area under the curve (AUC) values, nuclear UBE2I had the best diagnostic power (AUC = 1). Meanwhile, the knockdown of UBE2I can inhibit the proliferation of ccRCC cells. Conclusion: UBE2I, identified by CRISPR-cas9 screening, was a core gene-regulating ccRCC cell viability, which accumulated in the nucleus and acted as a potential novel promising diagnostic biomarker for ccRCC patients. Blocking the nuclear translocation of UBE2I may have potential therapeutic value with ccRCC patients.

AAN-Face: Attention Augmented Networks for Face Recognition.

Convolutional neural networks are capable of extracting powerful representations for face recognition. However, they tend to suffer from poor generalization due to imbalanced data distributions where a small number of classes are over-represented (e.g. frontal or non-occluded faces) and some of the remaining rarely appear (e.g. profile or heavily occluded faces). This is the reason why the performance is dramatically degraded in minority classes. For example, this issue is serious for recognizing masked faces in the scenario of ongoing pandemic of the COVID-19. In this work, we propose an Attention Augmented Network, called AAN-Face, to handle this issue. First, an attention erasing (AE) scheme is proposed to randomly erase units in attention maps. This well prepares models towards occlusions or pose variations. Second, an attention center loss (ACL) is proposed to learn a center for each attention map, so that the same attention map focuses on the same facial part. Consequently, discriminative facial regions are emphasized, while useless or noisy ones are suppressed. Third, the AE and the ACL are incorporated to form the AAN-Face. Since the discriminative parts are randomly removed by the AE, the ACL is encouraged to learn different attention centers, leading to the localization of diverse and complementary facial parts. Comprehensive experiments on various test datasets, especially on masked faces, demonstrate that our AAN-Face models outperform the state-of-the-art methods, showing the importance and effectiveness.

Adversarial Attack Against Deep Saliency Models Powered by Non-Redundant Priors.

Saliency detection is an effective front-end process to many security-related tasks, e.g. automatic drive and tracking. Adversarial attack serves as an efficient surrogate to evaluate the robustness of deep saliency models before they are deployed in real world. However, most of current adversarial attacks exploit the gradients spanning the entire image space to craft adversarial examples, ignoring the fact that natural images are high-dimensional and spatially over-redundant, thus causing expensive attack cost and poor perceptibility. To circumvent these issues, this paper builds an efficient bridge between the accessible partially-white-box source models and the unknown black-box target models. The proposed method includes two steps: 1) We design a new partially-white-box attack, which defines the cost function in the compact hidden space to punish a fraction of feature activations corresponding to the salient regions, instead of punishing every pixel spanning the entire dense output space. This partially-white-box attack reduces the redundancy of the adversarial perturbation. 2) We exploit the non-redundant perturbations from some source models as the prior cues, and use an iterative zeroth-order optimizer to compute the directional derivatives along the non-redundant prior directions, in order to estimate the actual gradient of the black-box target model. The non-redundant priors boost the update of some "critical" pixels locating at non-zero coordinates of the prior cues, while keeping other redundant pixels locating at the zero coordinates unaffected. Our method achieves the best tradeoff between attack ability and perturbation redundancy. Finally, we conduct a comprehensive experiment to test the robustness of 18 state-of-the-art deep saliency models against 16 malicious attacks, under both of white-box and black-box settings, which contributes a new robustness benchmark to the saliency community for the first time.

Dual-path Attention Network for Compressed Sensing Image Reconstruction.

Although deep neural network methods achieved much success in compressed sensing image reconstruction in recent years, they still have some issues, especially in preserving texture details. In this paper, we propose a new dual-path attention network for compressed sensing image reconstruction, which is composed of a structure path, a texture path and a texture attention module. Motivated by the classical paradigm of image structure-texture decomposition, the structure path aims to reconstruct the dominant structure component of the original image, and the texture path targets at recovering the remaining texture details. To better bridge the information between two paths, the texture attention module is designed to deliver the useful structure information to the texture path and predict the texture region, thereby facilitating the recovery of texture details. Two paths are optimized with a unified loss function. In the testing phase, given the measurement vector of a new image, it can be well reconstructed by carrying out the well trained dual-path attention network and integrating the outputs of the structure path and the texture path. Experimental results on the SET5, SET11 and BSD68 testing datasets demonstrate that the proposed method achieves comparable or better results compared with some state-of-the-art deep learning based methods and conventional iterative optimization based methods in terms of reconstruction quality and robustness to noise.

Task-Oriented Feature-Fused Network With Multivariate Dataset for Joint Face Analysis.

Deep multitask learning for face analysis has received increasing attentions. From literature, most existing methods focus on optimizing a main task by jointly learning several auxiliary tasks. It is challenging to consider the performance of each task in a multitask framework due to the following reasons: 1) different face tasks usually rely on different levels of semantic features; 2) each task has different learning convergence rate, which could affect the whole performance when joint training; and 3) multitask model needs rich label information for efficient training, but existing facial datasets provide limited annotations. To address these issues, we propose a task-oriented feature-fused network (TFN) for simultaneously solving face detection, landmark localization, and attribute analysis. In this network, a task-oriented feature-fused block is designed to learn task-specific feature combinations; then, an alternative multitask training scheme is presented to optimize each task with considering of their different learning capacities. We also present a large-scale face dataset called JFA in support of proposed method, which provides multivariate labels, including face bounding box, 68 facial landmarks, and 3 attribute labels (i.e., apparent age, gender, and ethnicity). The experimental results suggest that the TFN outperforms several multitask models on the JFA dataset. Furthermore, our approach achieves competitive performances on WIDER FACE and 300W dataset, and obtains state-of-the-art results for gender recognition on the MORPH II dataset.