Although Faster R-CNN and its variants have shown promising performance in object detection, they only exploit simple first-order representation of proposals for final classification regression. Recent methods demonstrate that the integration high-order statistics into deep convolutional neural networks can achieve impressive improvement, but their goal is to model whole images by discarding location information so cannot be directly adopted detection. In this paper, we make an attempt...

Recently, extraction of blood vessels has aroused widespread interests in medical image analysis. In this work, to accelerate convergence speed and enhance the representation for discriminative features, we introduce residual block structure ResNet into 3D U-Net, construct a new Residual U-Net architect segment hepatic portal veins from abdominal CT volumes. addition, develop weighted Dice loss function cope with challenges pixel imbalance, vessel boundary segmentation small segmentation....

Depth information opens up opportunities for video object segmentation (VOS) to be more accurate and robust in complex scenes. However, RGBD VOS is still unexplored due the high-cost collection time-consuming annotation of data. In this work, we first introduce a new benchmark VOS, named DepthVOS, which contains 350 videos (over 55k frames) annotated with masks bounding boxes. Then, propose novel strong baseline model - Fused Color-Depth Network (FusedCDNet) can learned merely under box...

This paper presents a factorization based active contour model for 2-phase texture segmentation. We utilize the local spectral histogram as features, and then establish novel energy function on theory of matrix decomposition. Unlike existing methods, we only choose combination weights from object region background to handle motion curve. compare proposed method recently methods experiments are performed synthetic real-world images. The experimental results show that our is more robust...

Automatic liver tumour segmentation is an important step towards digital medical research, clinical diagnosis and therapy planning. However, the existence of noise, low contrast heterogeneity make automatic remaining open challenge. In this work, we focus on a novel method to segment in abdomen images from CT scans using fully convolutional networks (FCN) non-negative matrix factorization (NMF) based deformable model. We train FCN for semantic preprocessed training data by BM3D. The...

Recent years have witnessed the prevalence of memory-based methods for Semi-supervised Video Object Segmentation (SVOS) which utilise past frames efficiently label propagation. When conducting feature matching, fine-grained multi-scale matching has typically been performed using all query points, inevitably results in redundant computations and thus makes fusion ineffective. In this paper, we develop a new Point-based Memory Network, termed as PMNet, to perform on hard samples only, assuming...

Referring Video Object Segmentation (RVOS) aims to segment the text-depicted object from video sequences. With excellent capabilities in long-range modelling and information interaction, transformers have been increasingly applied existing RVOS architectures. To better leverage multimodal data, most efforts focus on interaction between visual textual features. However, they ignore syntactic structures of text during where all components are intertwined, resulting ambiguous vision-language...

Race identification is an essential ability for human eyes. classification by machine based on face image can be used in some practical application fields. Employing holistic analysis, local feature extraction and 3D model, many race methods have been introduced. In this paper, we propose a novel fusion periocular region features classifying East Asian from Caucasian. With the landmarks, extract five textures or geometrical interesting regions which contain available discriminating...

Medical image segmentation plays an important role in digital medical research, and therapy planning delivery. However, the presence of noise low contrast renders automatic liver extremely challenging task. In this study, we focus on a variational approach to computed tomography scan volumes semiautomatic slice-by-slice manner. method, one slice is selected its connected component region determined manually initialize subsequent process. From guiding slice, execute proposed method downward...

Despite recent progress, Video Object Segmentation (VOS) remains challenging in complex situations such as low light and dark scenes. In this paper, we tackle the visibility limitations by introducing thermal information auxillary for VOS. Specifically, generate a hybrid benchmark dataset Visible-Thermal VOS, named VisT300, which contains 300 videos with visible frames corresponding object mask annotations. Besides, integration Network, VTiNet, is proposed to use both cross-modal cross-frame...

Images with weak contrast, overlapped noise and texture of the object background make many PDE based methods disabled. To address these problems, this paper presents a novel combined multi-scale variational framework level set segmentation model. Its formulation consists edge-based term, region-based term shape constraint term. The is constructed using newly defined edge stopping function. derived from parameter-free Gaussian probability density function (pdf) multiple kernel are used to...

Although Faster R-CNN and its variants have shown promising performance in object detection, they only exploit simple first-order representation of proposals for final classification regression. Recent methods demonstrate that the integration high-order statistics into deep convolutional neural networks can achieve impressive improvement, but their goal is to model whole images by discarding location information so cannot be directly adopted detection. In this paper, we make an attempt...

This paper introduces an effective active contour model for texture segmentation. To improve the robustness against noise and illumination, a novel descriptor named local statistical variation degree (LSVD) is presented to express textural features, which uses corner point deletion isolated region detection operations eliminate image patches unrelated with object regions. And then fused features combined LSVD Gabor can be constructed structure in many scene. During segmentation stage,...

Medical image segmentation plays an important role in digital medical research, therapy planning, and computer aided diagnosis. However, the existence of noise low contrast make automatic liver remains open challenge. In this work we focus on a novel variational semi-automatic method. First, used signed distance functions (SDF) representing pattern shapes to build statistical shape model. Then global Gaussian fitting energy enforced local feature were established guide PCA-based topological...

Active Shape Model (ASM) is a most effective method of facial landmarking. It employs two models, profile model and shape model, to match the position landmark. In this paper, we introduce new based on relative feature (RPF) in local region improve ASM. We found fact that landmarks with larger matching error have more displacement. So, our method, RPF used adjust displacement every iteration. STASM (Stacked ASM) practical standard ASM proved be best locating face landmarks. Our experiments...

Dougong is a unique culture in Chinese traditional architecture. In University, the Architectural students usually use video, pictures, and even handmade craft to learn knowledge about Dougong. However, making these complicated components by hands requires lot of facilities. To solve problems, this paper builds learning application using Virtual Reality (VR) technology, where can master how construct interacting with virtual models. addition module, creates simulated scene showing great...

Human evaluation is viewed as a reliable method for NLG which expensive and time-consuming. To save labor costs, researchers usually perform human on small subset of data sampled from the whole dataset in practice. However, different selection subsets will lead to rankings systems. give more correct inter-system ranking make gold standard reliable, we propose Constrained Active Sampling Framework (CASF) judgment. CASF operates through Learner, Systematic Sampler Controller select...

Human evaluation is viewed as a reliable method for NLG which expensive and time-consuming. To save labor costs, researchers usually perform human on small subset of data sampled from the whole dataset in practice. However, different selection subsets will lead to rankings systems. give more correct inter-system ranking make gold standard reliable, we propose Constrained Active Sampling Framework (CASF) judgment. CASF operates through Learner, Systematic Sampler Controller select...

The evaluation of natural language generation (NLG) tasks is a significant and longstanding research issue. With the recent emergence powerful large models (LLMs), some studies have turned to LLM-based automatic methods, which demonstrate great potential become new paradigm following traditional string-based model-based metrics. However, despite improved performance existing they still possess deficiencies, such as dependency on references limited flexibility. Therefore, in this paper, we...