Pseudo-label-based semi-supervised learning (SSL) has achieved great success on raw data utilization. However, its training procedure suffers from confirmation bias due to the noise contained in self-generated artificial labels. Moreover, model's judgment becomes noisier real-world applications with extensive out-of-distribution data. To address this issue, we propose a general method named Class-aware Contrastive Semi-Supervised Learning (CCSSL), which is drop-in helper improve pseudo-label...

Abstract Mitochondrial dysfunction contributes to the development of secondary brain injury (SBI) following intracerebral hemorrhage (ICH) and represents a promising therapeutic target. Celastrol, primary active component Tripterygium wilfordii , is natural product that exhibits mitochondrial neuronal protection in various cell types. This study aims investigate neuroprotective effects celastrol against ICH‐induced SBI explore its underlying mechanisms. Celastrol improves neurobehavioral...

Video anomaly detection aims to automatically identify unusual objects or behaviours by learning from normal videos. Previous methods tend use simplistic reconstruction prediction constraints, which leads the insufficiency of learned representations for data. As such, we propose a novel bi-directional architecture with three consistency constraints comprehensively regularize task pixel-wise, cross-modal, and temporal-sequence levels. First, predictive is proposed consider symmetry property...

The huge burden of computation and memory are two obstacles in ultra-high resolution image segmentation. To tackle these issues, most the previous works follow global-local refinement pipeline, which pays more attention to consumption but neglects inference speed. In comparison pipeline that partitions large into small local regions, we focus on inferring whole directly. this paper, propose ISDNet, a novel segmentation framework integrates shallow deep networks new manner, significantly...

The histogram equalization process is a simple yet efficient image contrast enhancement technique that generally produces satisfactory results. However, due to its design limitations, output images often experience loss of fine details or contain unwanted viewing artefacts. One reason for such imperfection failure some techniques fully utilize the allowable intensity range in conveying information captured from scene. proposed colour introduced this work aims at maximizing content within an...

Model generalization to the unseen scenes is crucial real-world applications, such as autonomous driving, which requires robust vision systems. To enhance model generalization, domain through learning domain-invariant representation has been widely studied. However, most existing works learn shared feature space within multi-source domains but ignore characteristic of itself (e.g., sensitivity domain-specific style). Therefore, we propose Domain-invariant Representation Learning (DIRL) for...

Weakly supervised object localization (WSOL) aims to learn localizer solely by using image-level labels. The convolution neural network (CNN) based techniques often result in highlighting the most discriminative part of objects while ignoring entire extent. Recently, transformer architecture has been deployed WSOL capture long-range feature dependencies with self-attention mechanism and multilayer perceptron structure. Nevertheless, transformers lack locality inductive bias inherent CNNs...

Parameter-efficient transfer learning (PETL) is an emerging research spot aimed at inexpensively adapting large-scale pre-trained models to downstream tasks. Recent advances have achieved great success in saving storage costs for various by updating a small number of parameters instead full tuning. However, we notice that most existing PETL methods still incur non-negligible latency during inference. In this paper, propose parameter-efficient and computational friendly adapter giant vision...

Text-driven 3D stylization is a complex and crucial task in the fields of computer vision (CV) graphics (CG), aimed at transforming bare mesh to fit tar-get text. Prior methods adopt text-independent multilayer perceptrons (MLPs) predict attributes target with supervision CLIP loss. However, such architecture lacks textual guidance during predicting attributes, thus leading unsatisfactory slow convergence. To address these limitations, we present X-Mesh, an innovative text-driven framework...

The enhancement of image contrast and preservation brightness are two important but conflicting objectives in restoration. Previous attempts based on linear histogram equalization had achieved enhancement, exact was not accomplished. A new perspective is taken here to provide balanced performance simultaneously by casting the quest such solution an optimization problem. Specifically, non-linear gamma correction method adopted enhance contrast, while a weighted sum approach employed for...

To bridge the ever-increasing gap between deep neural networks' complexity and hardware capability, network quantization has attracted more research attention. The latest trend of mixed precision takes advantage hardware's multiple bit-width arithmetic operations to unleash full potential quantization. However, existing approaches rely heavily on an extremely time-consuming search process various relaxations when seeking optimal bit configuration. address this issue, we propose optimize a...

Knowledge distillation (KD) has been widely explored in unsupervised anomaly detection (AD). The student is assumed to constantly produce representations of typical patterns within trained data, named "normality", and the representation discrepancy between teacher model identified as anomalies. However, it suffers from "normality forgetting" issue. Trained on anomaly-free still well reconstructs anomalous for anomalies sensitive fine normal which also appear training. To mitigate this issue,...

ABSTRACT Background Intracerebral hemorrhage (ICH) is a severe condition characterized by elevated mortality and disability rates. The cAMP‐activated exchange protein‐1 (EPAC‐1) implicated in various cytoprotective mechanisms; however, its specific role ICH remains unclear. Methods A rat model of was established injecting autologous blood, while the vitro primary neuronal stimulated using oxyhemoglobin (OxyHb). construction EPAC‐1 overexpression wild‐type (WT) phosphorylated mutant plasmids...

Recent Few-Shot Learning (FSL) methods put emphasis on generating a discriminative embedding features to precisely measure the similarity between support and query sets. Current CNN-based cross-attention approaches generate representations via enhancing mutually semantic similar regions of pairs. However, it suffers from two problems: CNN structure produces inaccurate attention map based local features, backgrounds cause distraction. To alleviate these problems, we design novel SpatialFormer...

Referring Expression Comprehension (REC) is a task of grounding the referent based on an expression, and its development greatly limited by expensive instance-level annotations. Most existing weakly supervised methods are built two-stage detection networks, which computationally expensive. In this paper, we resort to efficient one-stage detector propose novel model called RefCLIP.Specifically, RefCLIP redefines REC as anchor-text matching problem, can avoid complex post-processing in...

Pseudo-labeling is significant for semi-supervised instance segmentation, which generates masks and classes from unannotated images subsequent training. However, in existing pipelines, pseudo-labels that contain valuable information may be directly filtered out due to mismatches class mask quality. To address this issue, we propose a novel framework, called pseudo-label aligning segmentation (PAIS), paper. In PAIS, devise dynamic loss (DALoss) adjusts the weights of terms with varying score...

Referring expression comprehension (REC) often requires a large number of instance-level annotations for fully supervised learning, which are laborious and expensive. In this paper, we present the first attempt semi-supervised learning REC propose strong baseline method called RefTeacher. Inspired by recent progress in computer vision, RefTeacher adopts teacher-student paradigm, where teacher network predicts pseudolabels optimizing student one. This paradigm allows models to exploit massive...

Text-based person retrieval (TPR) is a challenging task that involves retrieving specific individual based on textual description. Despite considerable efforts to bridge the gap between vision and language, significant differences these modalities continue pose challenge. Previous methods have attempted align text image samples in modal-shared space, but they face uncertainties optimization directions due movable features of both failure account for one-to-many relationships image-text pairs...

Interactive image segmentation enables annotators to efficiently perform pixel-level annotation for tasks. However, the existing interactive pipeline suffers from inefficient computations of models because following two issues. First, annotators' later click is based on models' feedback former click. This serial interaction unable utilize model's parallelism capabilities. Second, in each step, model handles invariant along with sparse variable clicks, resulting a process that's highly...

Posterior circulation aneurysms have been regarded as the most challenging for endovascular coiling and microsurgical occlusion. The role of treatment is gradually being overlooked diminishing in trend treatment. As occlusion posterior decreasing, we present our relevant experience to evaluate options surgical approaches. A retrospective study was conducted Department Neurosurgery First Affiliated Hospital Soochow University between 2016 2021. Patients with treated by clipping, bypass,...