Preterm birth associated alterations in brain structure, cognitive functioning and behavior in children from the ABCD dataset.

BACKGROUND: Preterm birth is a global health problem and associated with increased risk of long-term developmental impairments, but findings on the adverse outcomes of prematurity have been inconsistent. METHODS: Data were obtained from the baseline session of the ongoing longitudinal Adolescent Brain and Cognitive Development (ABCD) Study. We identified 1706 preterm children and 1865 matched individuals as Control group and compared brain structure (MRI data), cognitive function and mental health symptoms. RESULTS: Results showed that preterm children had higher psychopathological risk and lower cognitive function scores compared to controls. Structural MRI analysis indicated that preterm children had higher cortical thickness in the medial orbitofrontal cortex, parahippocampal gyrus, temporal and occipital gyrus; smaller volumes in the temporal and parietal gyrus, cerebellum, insula and thalamus; and smaller fiber tract volumes in the fornix and parahippocampal-cingulum bundle. Partial correlation analyses showed that gestational age and birth weight were associated with ADHD symptoms, picvocab, flanker, reading, fluid cognition composite, crystallized cognition composite and total cognition composite scores, and measures of brain structure in regions involved with emotional regulation, attention and cognition. CONCLUSIONS: These findings suggest a complex interplay between psychopathological risk and cognitive deficits in preterm children that is associated with changes in regional brain volumes, cortical thickness, and structural connectivity among cortical and limbic brain regions critical for cognition and emotional well-being.

Associations between body mass index, sleep-disordered breathing, brain structure, and behavior in healthy children.

Pediatric overweight/obesity can lead to sleep-disordered breathing (SDB), abnormal neurological and cognitive development, and psychiatric problems, but the associations and interactions between these factors have not been fully explored. Therefore, we investigated the associations between body mass index (BMI), SDB, psychiatric and cognitive measures, and brain morphometry in 8484 children 9-11 years old using the Adolescent Brain Cognitive Development dataset. BMI was positively associated with SDB, and both were negatively correlated with cortical thickness in lingual gyrus and lateral orbitofrontal cortex, and cortical volumes in postcentral gyrus, precentral gyrus, precuneus, superior parietal lobule, and insula. Mediation analysis showed that SDB partially mediated the effect of overweight/obesity on these brain regions. Dimensional psychopathology (including aggressive behavior and externalizing problem) and cognitive function were correlated with BMI and SDB. SDB and cortical volumes in precentral gyrus and insula mediated the correlations between BMI and externalizing problem and matrix reasoning ability. Comparisons by sex showed that obesity and SDB had a greater impact on brain measures, cognitive function, and mental health in girls than in boys. These findings suggest that preventing childhood obesity will help decrease SDB symptom burden, abnormal neurological and cognitive development, and psychiatric problems.

Associations among body mass index, working memory performance, gray matter volume, and brain activation in healthy children.

To investigate the neural mechanisms underlying the association between poorer working memory performance and higher body mass index (BMI) in children. We employed structural-(sMRI) and functional magnetic resonance imaging (fMRI) with a 2-back working memory task to examine brain abnormalities and their associations with BMI and working memory performance in 232 children with overweight/obesity (OW/OB) and 244 normal weight children (NW) from the Adolescent Brain Cognitive Development dataset. OW/OB had lower working memory accuracy, which was associated with higher BMI. They showed smaller gray matter (GM) volumes in the left superior frontal gyrus (SFG_L), dorsal anterior cingulate cortex, medial orbital frontal cortex, and medial superior frontal gyrus, which were associated with lower working memory accuracy. During the working memory task, OW/OB relative to NW showed weaker activation in the left superior temporal pole, amygdala, insula, and bilateral caudate. In addition, caudate activation mediated the relationship between higher BMI and lower working memory accuracy. Higher BMI is associated with smaller GM volumes and weaker brain activation in regions involved with working memory. Task-related caudate dysfunction may account for lower working memory accuracy in children with higher BMI.

Obesity is associated with decreased gray matter volume in children: a longitudinal study.

Childhood obesity has become a global health problem. Previous studies showed that childhood obesity is associated with brain structural differences relative to controls. However, few studies have been performed with longitudinal evaluations of brain structural developmental trajectories in childhood obesity. We employed voxel-based morphometry (VBM) analysis to assess gray matter (GM) volume at baseline and 2-year follow-up in 258 obese children (OB) and 265 normal weight children (NW), recruited as part of the National Institutes of Health Adolescent Brain and Cognitive Development study. Significant group × time effects on GM volume were observed in the prefrontal lobe, thalamus, right precentral gyrus, caudate, and parahippocampal gyrus/amygdala. OB compared with NW had greater reductions in GM volume in these regions over the 2-year period. Body mass index (BMI) was negatively correlated with GM volume in prefrontal lobe and with matrix reasoning ability at baseline and 2-year follow-up. In OB, Picture Test was positively correlated with GM volume in the left orbital region of the inferior frontal gyrus (OFCinf_L) at baseline and was negatively correlated with reductions in OFCinf_L volume (2-year follow-up vs. baseline). These findings indicate that childhood obesity is associated with GM volume reduction in regions involved with reward evaluation, executive function, and cognitive performance.

Targeting VEGF-A/VEGFR2 Y949 Signaling-Mediated Vascular Permeability Alleviates Hypoxic Pulmonary Hypertension.

BACKGROUND: Pulmonary hypertension (PH) is associated with increased expression of VEGF-A (vascular endothelial growth factor A) and its receptor, VEGFR2 (vascular endothelial growth factor 2), but whether and how activation of VEGF-A signal participates in the pathogenesis of PH is unclear. METHODS: VEGF-A/VEGFR2 signal activation and VEGFR2 Y949-dependent vascular leak were investigated in lung samples from patients with PH and mice exposed to hypoxia. To study their mechanistic roles in hypoxic PH, we examined right ventricle systolic pressure, right ventricular hypertrophy, and pulmonary vasculopathy in mutant mice carrying knock-in of phenylalanine that replaced the tyrosine at residual 949 of VEGFR2 (Vefgr2Y949F) and mice with conditional endothelial deletion of Vegfr2 after chronic hypoxia exposure. RESULTS: We show that PH leads to excessive pulmonary vascular leak in both patients and hypoxic mice, and this is because of an overactivated VEGF-A/VEGFR2 Y949 signaling axis. In the context of hypoxic PH, activation of Yes1 and c-Src and subsequent VE-cadherin phosphorylation in endothelial cells are involved in VEGFR2 Y949-induced vascular permeability. Abolishing VEGFR2 Y949 signaling by Vefgr2Y949F point mutation was sufficient to prevent pulmonary vascular permeability and inhibit macrophage infiltration and Rac1 activation in smooth muscle cells under hypoxia exposure, thereby leading to alleviated PH manifestations, including muscularization of distal pulmonary arterioles, elevated right ventricle systolic pressure, and right ventricular hypertrophy. It is important that we found that VEGFR2 Y949 signaling in myeloid cells including macrophages was trivial and dispensable for hypoxia-induced vascular abnormalities and PH. In contrast with selective blockage of VEGFR2 Y949 signaling, disruption of the entire VEGFR2 signaling by conditional endothelial deletion of Vegfr2 promotes the development of PH. CONCLUSIONS: Our results support the notion that VEGF-A/VEGFR2 Y949-dependent vascular permeability is an important determinant in the pathogenesis of PH and might serve as an attractive therapeutic target pathway for this disease.

A single-cell transcriptomic atlas of the human ciliary body.

The ciliary body critically contributes to the ocular physiology with multiple responsibilities in the production of aqueous humor, vision accommodation and intraocular immunity. Comparatively little work, however, has revealed the single-cell molecular taxonomy of the human ciliary body required for studying these functionalities. In this study, we report a comprehensive atlas of the cellular and molecular components of human ciliary body as well as their interactions using single-cell RNA sequencing (scRNAseq). Cluster analysis of the transcriptome of 14,563 individual ciliary cells from the eyes of 3 human donors identified 14 distinct cell types, including the ciliary epithelium, smooth muscle, vascular endothelial cell, immune cell and other stromal cell populations. Cell-type discriminative gene markers were also revealed. Unique gene expression patterns essential for ciliary epithelium-mediated aqueous humor inflow and ciliary smooth muscle contractility were identified. Importantly, we discovered the transitional states that probably contribute to the transition of ciliary macrophage into retina microglia and verified no lymphatics in the ciliary body. Moreover, the utilization of CellPhoneDB allowed us to systemically infer cell-cell interactions among diverse ciliary cells including those that potentially participate in the pathogenesis of glaucoma and uveitis. Altogether, these new findings provide insights into the regulation of intraocular pressure, accommodation reflex and immune homeostasis under physiological and pathological conditions.

Classification of Visual Field Abnormalities in Highly Myopic Eyes without Pathologic Change.

PURPOSE: To develop a classification system of visual field (VF) abnormalities in highly myopic eyes with and without glaucoma. DESIGN: Secondary analysis of VF data from a longitudinal cohort study. PARTICIPANTS: One thousand eight hundred ninety-three VF tests from 1302 eyes (825 individuals). METHODS: All participants underwent VF testing (Humphrey 24-2 Swedish interactive threshold algorithm standard program; Carl Zeiss Meditec) and detailed ophthalmic examination. A comprehensive set of VF defect patterns was defined via observation of the 1893 VF reports, literature review, and consensus meetings. The classification system comprised 4 major types of VF patterns, including normal type, glaucoma-like defects (paracentral defect, nasal step, partial arcuate defect, arcuate defect), high myopia-related defects (enlarged blind spot, vertical step, partial peripheral rim, nonspecific defect), and combined defects (nasal step with enlarged blind spot). A subset (n = 1000) of the VFs was used to evaluate the interobserver and intraobserver agreement and weighted κ values of the classification system by 2 trained readers. The prevalence of various VF patterns and their associated factors were determined. MAIN OUTCOME MEASURES: The classification of VF in highly myopic eyes and its associated risk factors. RESULTS: We found that normal type, glaucoma-like defects, high myopia-related defects, and combined defects accounted for 74.1%, 10.8%, 15.0%, and 0.1% of all unique VF tests, respectively. The interobserver and intraobserver agreements were > 89%, and the corresponding κ values were 0.86 or more between readers. Both glaucoma-like and high myopia-related VF defects were associated with older age (odds ratios [ORs], 1.07 [95% confidence interval (CI), 1.04-1.10; P < 0.001] and 1.06 [95% CI, 1.04-1.10; P < 0.001]) and longer axial length (ORs, 1.65 [95% CI, 1.32-2.07; P < 0.001] and 1.37 [95% CI, 1.11-1.68; P = 0.003]). Longer axial length showed a stronger effect on the prevalence of glaucoma-like VF defects than on the prevalence of high myopia-related VF defects (P = 0.036). CONCLUSIONS: We propose a new and reproducible classification system of VF abnormalities for nonpathologic high myopia. Applying a comprehensive classification system will facilitate communication and comparison of findings among studies.

Reconstructing seen image from brain activity by visually-guided cognitive representation and adversarial learning.

Reconstructing perceived stimulus (image) only from human brain activity measured with functional Magnetic Resonance Imaging (fMRI) is a significant task in brain decoding. However, the inconsistent distribution and representation between fMRI signals and visual images cause great 'domain gap'. Moreover, the limited fMRI data instances generally suffer from the issues of low signal noise ratio (SNR), extremely high dimensionality, and limited spatial resolution. Existing methods are often affected by these issues so that a satisfactory reconstruction is still an open problem. In this paper, we show that it is possible to obtain a promising solution by learning visually-guided latent cognitive representations from the fMRI signals, and inversely decoding them to the image stimuli. The resulting framework is called Dual-Variational Autoencoder/ Generative Adversarial Network (D-Vae/Gan), which combines the advantages of adversarial representation learning with knowledge distillation. In addition, we introduce a novel three-stage learning strategy which enables the (cognitive) encoder to gradually distill useful knowledge from the paired (visual) encoder during the learning process. Extensive experimental results on both artificial and natural images have demonstrated that our method could achieve surprisingly good results and outperform the available alternatives.

A Comparison of Subconjunctival Wound Healing between Different Methods of Dissecting Subconjunctival Tissues.

OBJECTIVE: To compare different methods for dissecting subconjunctival tissues by developing subconjunctival wound healing models. METHODS: New Zealand white rabbits were separated into 3 groups based on the method by which the rabbit subconjunctival wound healing model was generated: subconjunctival tissues were dissected episclerally (EPI) or subepithelially (SUB), with a corresponding blank control (CON). All the cases in the experimental groups were surgically prepared with conjunctival flaps, and they were sacrificed on the third postoperative day. At the surgical sites, the protein levels of hypoxia-inducible factor-1 (HIF-1)-α, vascular endothelial growth factor (VEGF)-A, and matrix metalloproteinase (MMP)-2 were detected by Western blot, morphological vascularity was measured by Adobe Photoshop, and subconjunctival fibrosis was assessed by histology. RESULTS: Compared with the CON group, both the EPI and SUB groups showed significantly upregulated protein levels of HIF-1α, VEGF-A, and MMP-2. In addition, the protein levels of HIF-1α, VEGF-A, and MMP-2 were higher in the EPI group than in the SUB group. Morphological vascularity was significantly elevated in the EPI group compared with the SUB and CON groups. Collagen content was markedly increased in the EPI group compared with the SUB and CON groups. CONCLUSIONS: Dissecting subconjunctival tissues subepithelially inhibits subconjunctival fibrosis, which may be instructive in tenonectomy in filtration surgery.

Bacterial agents and changes in drug susceptibilities in cases of chronic dacryocystitis, Southern China.

PURPOSE: This study aimed to determine the susceptibility and the changes of bacterial agents of chronic dacryocystitis and determine the risk factors for bacterial prevalence and drug sensitivity to provide a reference for clinical selection of antibiotics. METHODS: A case-control study was conducted using 112 patients with chronic dacryocystitis and 112 patients with non-infectious ophthalmopathy between August 2017 and April 2018. Lacrimal and conjunctival sac secretions were cultured for aerobic and anaerobic bacteria. Forty-five patients with chronic dacryocystitis between November 2014 and November 2015 were also included. RESULTS: Positive bacterial cultures were obtained from 61.9% and 50.9% of chronic dacryocystitis and non-infectious ophthalmopathy patients, but the detection rates for pathogenic bacteria were 18.3% and 2.7%, respectively (P > 0.001). Gram-negative and anaerobic bacteria were significantly more prevalent in the patient group compared with the control group (P = 0.001 and 0.005, respectively). Bacteria were detected at a significantly higher rate in patients with irritant symptoms (itch or foreign-body sensation) than in those without (OR = 9.333, P = 0.002), particularly Staphylococcus (OR = 9.783, P = 0.002). 11.6% (10/86) and 55.8% (48/86) showed resistance to levofloxacin and tobramycin, respectively. Compared with three years ago, the detection rate for Gram-positive cocci decreased from 51.1% to 27.8% (χ2 = 8.054, P = 0.005) CONCLUSIONS: Gram-positive cocci, Gram-negative bacilli, and anaerobic bacteria were the predominant pathogens. The prevalence of Gram-positive bacteria in cases of chronic dacryocystitis is decreasing.

Gut microbiota compositional profile and serum metabolic phenotype in patients with primary open-angle glaucoma.

The gut microbiota (GM) and its influence on host metabolism are considered to be an environmental factor that contributes to the progression of many immune and neurodegenerative diseases. However, the features of the GM and serum metabolites in Primary open-angle glaucoma (POAG) patients have not been clearly elucidated. The purpose of this research is to explore the gut microbial composition and serum metabolic phenotype in POAG patients. 16S rRNA V4 genes of bacteria from the fecal samples of 30 POAG patients and 30 healthy subjects were sequenced by the Illumina MiSeq platform and then analyzed by QIIME. Their serum samples were analyzed by gas chromatography/mass spectrometry (GC-MS)-based metabolomics. The association between gut microbial species and host circulating metabolites and clinical phenotypes was also analyzed. Compared with controls, f Prevotellaceae, g unidentified Enterobacteriaceae, and s Escherichia coli increased the most in POAG patients, whereas g Megamonas and s Bacteroides plebeius significantly decreased in POAG patients. The alteration of the endogenous metabolomic profile in POAG patients included five amino acids or dipeptides, two hormone derivates, one purine derivative, one bile acid derivative and one organic acid. It also showed that citric acid was positively correlated with Megamonas, whereas L-γ-Glutamyl-L-alanine, MHPG, cholic acid glucuronide and hypoxanthine were negatively correlated with Megamonas. Mean visual acuity was negatively correlated with Blautia, mean VF-MD was negatively correlated with Faecalibacterium, and average RNFL thickness was positively correlated with Streptococcus. Our results revealed that there was a distinct difference in GM composition and serum metabolic phenotype between POAG patients and healthy individuals. This finding suggests the potential correlations between the GM and serum metabolites in the pathogenesis of glaucoma and thus provides new insight into the GM-targeted interventions of this disease.

Inter-site harmonization based on dual generative adversarial networks for diffusion tensor imaging: application to neonatal white matter development.

BACKGROUND: Site-specific variations are challenges for pooling analyses in multi-center studies. This work aims to propose an inter-site harmonization method based on dual generative adversarial networks (GANs) for diffusion tensor imaging (DTI) derived metrics on neonatal brains. RESULTS: DTI-derived metrics (fractional anisotropy, FA; mean diffusivity, MD) are obtained on age-matched neonates without magnetic resonance imaging (MRI) abnormalities: 42 neonates from site 1 and 42 neonates from site 2. Significant inter-site differences of FA can be observed. The proposed harmonization approach and three conventional methods (the global-wise scaling, the voxel-wise scaling, and the ComBat) are performed on DTI-derived metrics from two sites. During the tract-based spatial statistics, inter-site differences can be removed by the proposed dual GANs method, the voxel-wise scaling, and the ComBat. Among these methods, the proposed method holds the lowest median values in absolute errors and root mean square errors. During the pooling analysis of two sites, Pearson correlation coefficients between FA and the postmenstrual age after harmonization are larger than those before harmonization. The effect sizes (Cohen's d between males and females) are also maintained by the harmonization procedure. CONCLUSIONS: The proposed dual GANs-based harmonization method is effective to harmonize neonatal DTI-derived metrics from different sites. Results in this study further suggest that the GANs-based harmonization is a feasible pre-processing method for pooling analyses in multi-center studies.

ASI aurora search: an attempt of intelligent image processing for circular fisheye lens.

The circular fisheye lens exhibits an approximately 180° angular field-of-view (FOV), which is much larger than that of an ordinary lens. Thus, images captured with a circular fisheye lens are distributed non-uniformly with spherical deformation. Along with the fast development of deep neural networks for normal images, how to apply it to achieve intelligent image processing for a circular fisheye lens is a new task of significant importance. In this paper, we take the aurora images captured with all-sky-imagers (ASI) as a typical example. By analyzing the imaging principle of ASI and the magnetic characteristics of the aurora, a deformed region division (DRD) scheme is proposed to replace the region proposals network (RPN) in the advanced mask regional convolutional neural network (Mask R-CNN) framework. Thus, each image can be regarded as a "bag" of deformed regions represented with CNN features. After clustering all CNN features to generate a vocabulary, each deformed region is quantified to its nearest center for indexing. On the stage of an online search, a similarity score is computed by measuring the distances between regions in the query image and all regions in the data set, and the image with the highest value is outputted as the top rank search result. Experimental results show that the proposed method greatly improves the search accuracy and efficiency, demonstrating that it is a valuable attempt of intelligent image processing for circular fisheye lenses.

Limited versus total epithelial debridement ocular surface injury: Live fluorescence imaging of hemangiogenesis and lymphangiogenesis in Prox1-GFP/Flk1::Myr-mCherry mice.

BACKGROUND: Immunohistochemical staining experiments have shown that both hemangiogenesis and lymphangiogenesis occur following severe corneal and conjunctival injury and that the neovascularization of the cornea often has severe visual consequences. To better understand how hemangiogenesis and lymphangiogenesis are induced by different degrees of ocular injury, we investigated patterns of injury-induced corneal neovascularization in live Prox1-GFP/Flk1::myr-mCherry mice, in which blood and lymphatic vessels can be imaged simultaneously in vivo. METHODS: The eyes of Prox1-GFP/Flk1::myr-mCherry mice were injured according to four models based on epithelial debridement of the: A) central cornea (a 1.5-mm-diameter circle of tissue over the corneal apex), B) total cornea, C) bulbar conjunctiva, and D) cornea+bulbar conjunctiva. Corneal blood and lymphatic vessels were imaged on days 0, 3, 7, and 10 post-injury, and the percentages of the cornea containing blood and lymphatic vessels were calculated. RESULTS: Neither central corneal nor bulbar conjunctival debridement resulted in significant vessel growth in the mouse cornea, whereas total corneal and corneal+bulbar conjunctival debridement did. On day 10 in the central cornea, total cornea, bulbar conjunctiva, and corneal+bulbar conjunctival epithelial debridement models, the percentage of the corneal surface that was occupied by blood vessels (hemangiogenesis) was 1.9±0.8%, 7.14±2.4%, 2.29±1%, and 15.05±2.14%, respectively, and the percentage of the corneal surface that was occupied by lymphatic vessels (lymphangiogenesis) was 2.45±1.51%, 4.85±0.95%, 2.95±1.27%, and 4.15±3.85%, respectively. CONCLUSIONS: Substantial corneal debridement was required to induce corneal neovascularization in the mouse cornea, and the corneal epithelium may therefore be partially responsible for maintaining corneal avascularity. GENERAL SIGNIFICANCE: Our study demonstrates that GFP/Flk1::myr-mCherry mice are a useful model for studying coordinated hemangiogenic and lymphangiogenic responses.

Prox1-GFP/Flt1-DsRed transgenic mice: an animal model for simultaneous live imaging of angiogenesis and lymphangiogenesis.

The roles of angiogenesis in development, health, and disease have been studied extensively; however, the studies related to lymphatic system are limited due to the difficulty in observing colorless lymphatic vessels. But recently, with the improved technique, the relative importance of the lymphatic system is just being revealed. We bred transgenic mice in which lymphatic endothelial cells express GFP (Prox1-GFP) with mice in which vascular endothelial cells express DsRed (Flt1-DsRed) to generate Prox1-GFP/Flt1-DsRed (PGFD) mice. The inherent fluorescence of blood and lymphatic vessels allows for direct visualization of blood and lymphatic vessels in various organs via confocal and two-photon microscopy and the formation, branching, and regression of both vessel types in the same live mouse cornea throughout an experimental time course. PGFD mice were bred with CDh5CreERT2 and VEGFR2lox knockout mice to examine specific knockouts. These studies showed a novel role for vascular endothelial cell VEGFR2 in regulating VEGFC-induced corneal lymphangiogenesis. Conditional deletion of vascular endothelial VEGFR2 abolished VEGFA- and VEGFC-induced corneal lymphangiogenesis. These results demonstrate the potential use of the PGFD mouse as a powerful animal model for studying angiogenesis and lymphangiogenesis.

Swept-source optical coherence tomography imaging of macular retinal and choroidal structures in healthy eyes.

BACKGROUND: To report the thickness of the retina, retinal ganglion cell (RGC)-related layers, and choroid in healthy subjects using swept source optical coherence tomography (SS-OCT). METHODS: One hundred and forty-six healthy volunteers were consecutively recruited for this prospective observational study. Thickness of retina, RGC-related layers, and choroid in the standard early treatment of diabetic retinopathy study (ETDRS) grid were automatically measured using one SS-OCT (DRI OCT-1, Topcon, Japan). The IOL Master (Carl Zeiss Meditec, Germany) was used to measure axial length (AL). RESULTS: Thicknesses of the average macular ganglion cell complex (GCC) and ganglion cell-inner plexiform layer (GCIPL) were 105.3±9.7 and 78.5±6.2 um respectively. Neither of them was significantly related with sex, age, or AL. Both showed strong correlations with retinal thickness (r=0.793, p=0.000; r=0.813, p=0.000, respectively) and with similar topographic distributions within the retina. The thicknesses of retina and GCC/GCIPL in the inner sectors were significantly higher than in the outer sectors of the EDTRS area, while in the same region of the macula, the choroid exhibited completely different patterns of topographic variation. Men had 7.8 um thicker retina and 34.9 um thicker choroid than women after adjustment for age and AL (all p<0.05). Age and AL could significantly influence the choroidal thickness but not the retina (all p<0.05). CONCLUSION: Thickness of GCC/GCIPL in healthy Chinese individuals is not dramatically different across gender, age, and AL groups in terms of ETDRS grid, but sex is critical for retinal and choroidal thickness. Choroidal structure (but not retinal) can be significantly influenced by age and AL.

Semi-Supervised Learning With Heterogeneous Distribution Consistency for Visible Infrared Person Re-Identification.

Visible infrared person re-identification (VI-ReID) exposes considerable challenges because of the modality gaps between the person images captured by daytime visible cameras and nighttime infrared cameras. Several fully-supervised VI-ReID methods have improved the performance with extensive labeled heterogeneous images. However, the identity of the person is difficult to obtain in real-world situations, especially at night. Limited known identities and large modality discrepancies impede the effectiveness of the model to a great extent. In this paper, we propose a novel Semi-Supervised Learning framework with Heterogeneous Distribution Consistency (HDC-SSL) for VI-ReID. Specifically, through investigating the confidence distribution of heterogeneous images, we introduce a Gaussian Mixture Model-based Pseudo Labeling (GMM-PL) method, which adaptively adjusts different thresholds for each modality to label the identity. Moreover, to facilitate the representation learning of unutilized data whose prediction is lower than the threshold, Modality Consistency Regularization (MCR) is proposed to ensure the prediction consistency of the cross-modality pedestrian images and handle the modality variance. Extensive experiments with different label settings on two VI-ReID datasets demonstrate the effectiveness of our method. Particularly, HDC-SSL achieves competitive performance with state-of-the-art fully-supervised VI-ReID methods on RegDB dataset with only 1 visible label and 1 infrared label per class.

BPMTrack: Multi-Object Tracking With Detection Box Application Pattern Mining.

The key to multi-object tracking is its stability and the retention of identity information. A common problem with most detection-based approaches is trusting and using all the detector outputs for the association. However, some settings of detectors can affect stable long-range tracking. Based on the principle of reducing the association noise in the detection processing step, we propose a new framework, the Box application Pattern Mining Tracker (BPMTrack), to address this issue. Specifically, we worked on three main aspects: output threshold, association strategy, and motion model. Due to the problem of inconsistency between classification scores and localization accuracy, we propose the Box Quality Estimation Network (BQENet) to predict the localization quality scores of all detections in the current frame, reserving high-quality boxes for the tracker. In addition, based on observations of intensive scenarios, we propose a simple and effective data association method, the Non-Maximum Suppression Integration (NMSI) matching strategy. It recovers the Non-Maximum Suppression (NMS) detection, inputs them into BQENet, and then performs hierarchical matching with reasonable control of box priority to alleviate the problem of absent objects caused by occlusion. Finally, we propose an improved Measurement Correct and Noise Scale (MCNS) Kalman algorithm to improve the prediction accuracy of object positions and, thus, the association quality. We performed an extensive ablation evaluation of the proposed framework to prove its effectiveness. Moreover, the three tracking benchmarks show our method's accuracy and long-distance performance.

Few-Shot Face Stylization via GAN Prior Distillation.

Face stylization has made notable progress in recent years. However, when training on limited data, the performance of existing approaches significantly declines. Although some studies have attempted to tackle this problem, they either failed to achieve the few-shot setting (less than 10) or can only get suboptimal results. In this article, we propose GAN Prior Distillation (GPD) to enable effective few-shot face stylization. GPD contains two models: a teacher network with GAN Prior and a student network that fulfills end-to-end translation. Specifically, we adapt the teacher network trained on large-scale data in the source domain to the target domain using a handful of samples, where it can learn the target domain's knowledge. Then, we can achieve few-shot augmentation by generating source domain and target domain images simultaneously with the same latent codes. We propose an anchor-based knowledge distillation module that can fully use the difference between the training and the augmented data to distill the knowledge of the teacher network into the student network. The trained student network achieves excellent generalization performance with the absorption of additional knowledge. Qualitative and quantitative experiments demonstrate that our method achieves superior results than state-of-the-art approaches in a few-shot setting.

Meta Invariance Defense Towards Generalizable Robustness to Unknown Adversarial Attacks.

Despite providing high-performance solutions for computer vision tasks, the deep neural network (DNN) model has been proved to be extremely vulnerable to adversarial attacks. Current defense mainly focuses on the known attacks, but the adversarial robustness to the unknown attacks is seriously overlooked. Besides, commonly used adaptive learning and fine-tuning technique is unsuitable for adversarial defense since it is essentially a zero-shot problem when deployed. Thus, to tackle this challenge, we propose an attack-agnostic defense method named Meta Invariance Defense (MID). Specifically, various combinations of adversarial attacks are randomly sampled from a manually constructed Attacker Pool to constitute different defense tasks against unknown attacks, in which a student encoder is supervised by multi-consistency distillation to learn the attack-invariant features via a meta principle. The proposed MID has two merits: 1) Full distillation from pixel-, feature- and prediction-level between benign and adversarial samples facilitates the discovery of attack-invariance. 2) The model simultaneously achieves robustness to the imperceptible adversarial perturbations in high-level image classification and attack-suppression in low-level robust image regeneration. Theoretical and empirical studies on numerous benchmarks such as ImageNet verify the generalizable robustness and superiority of MID under various attacks.