Weakly supervised crowd counting involves the regression of number individuals present in an image, using only total as label. However, this task is plagued by two primary challenges: large variation head size and uneven distribution density. To address these issues, we propose a novel Hypergraph Association Crowd Counting (HACC) framework. Our approach consists new multi-scale dilated pyramid module that can efficiently handle size. Further, hypergraph association to solve problem density...

Most existing weakly supervised crowd counting methods utilize Convolutional Neural Networks (CNN) or Transformer to estimate the total number of individuals in an image. However, both CNN-based (grid-to-count paradigm) and Transformer-based (sequence-to-count take images as inputs a regular form. This approach treats all pixels equally but cannot address uneven distribution problem within human crowds. challenge would lead decline performance model. Compared with grid sequence, graph...

Abstract Crowd counting provides an important foundation for public security and urban management. Due to the existence of small targets large density variations in crowd images, is a challenging task. Mainstream methods usually apply convolution neural networks (CNNs) regress map, which requires annotations individual persons counts. Weakly-supervised can avoid detailed labeling only require counts as but existing fail achieve satisfactory performance because global perspective field...

Although deep networks based methods outperform traditional 3D reconstruction which require multiocular images or class labels to recover the full geometry, they may produce incomplete recovery and unfaithful when facing occluded parts of objects. To address these issues, we propose Depth-preserving Latent Generative Adversarial Network (DLGAN) consists Encoder-Decoder GAN (EDGAN, serving as a generator discriminator) Extreme Learning Machine (ELM, classifier) for from monocular depth image...

We propose two-dimensional pose estimation from a single range image of the human body, using sparse regression with componentwise clustering feature point representation (CCFPR) model. CCFPR includes primary points and secondary points. The consist torso center five extremal further serve to classify all body pixels as six components. are given by cluster centers each components other than torso, K-means cluster. is obtained learning projection matrix, which maps skeleton based on...

Pose tracking from range image sequences remains a difficult task due to strong noise and serious self-occlusion of human body. Existing work either rely on extremely large precisely annotated datasets, or accurate mesh model GPU acceleration. In this paper, we propose an unsupervised real-time framework pose sequences. Our consists visible hybrid (VHM), componentwise correspondence optimization (CCO) dynamic database lookup (DDL). VHM component sphere sets spherical point which exhibits...

Recovering the geometry of an object from a single depth image is interesting yet challenging problem. While previous learning based approaches have demonstrated promising performance, they don’t fully explore spatial relationships objects, which leads to unfaithful and incomplete 3D reconstruction. To address these issues, we propose Spatial Relationship Preserving Adversarial Network (SRPAN) consisting Capsule Attention Generative (3DCAGAN) 2D (2DGAN) for coarse-to-fine reconstruction view...

Human poses admit complicated articulations and multigranular similarity. Previous works on learning human pose metric utilize sparse models, which concentrate large weights highly similar fail to depict an overall structure of with Moreover, previous require a number similar/dissimilar annotated pairwise poses, is tedious task remains inaccurate due different subjective judgments experts. Motivated by graph-based neighbor assignment techniques, we propose unsupervised model called sparsity...

Abstract Traffic forecasting is an important part in realising intelligent traffic management, which helps controllers and travellers make effective decisions. However, accuracy often affected by missing data due to hardware software failure. Therefore, accurate prediction based on incomplete problem as well a challenge. Though many approaches recover the values before prediction, errors from data‐filling step are likely cause additional bias result. Besides, this tactic difficult guarantee...

Abstract This paper presents a Chinese Sign Language Markup (CSLML), which is developed for expressive sign language synthesis by introducing features and structure of prosody. The tags CSLML are divided into two levels: function level phonetic level. Function provides abstract information about signed content prosody, so it facilitates text annotating text‐driven automatic synthetic system adapts to diversified methods, such as motion capture animation or image‐based synthesis, may not be...

Image inpainting is a challenging task in image processing and widely applied many areas such as photo editing. Traditional patch-based methods are not effective to deal with complex or non-repetitive structures. Recently, deep learning-based approaches have shown promising results for inpainting. However, they usually generate contents artificial boundaries, distorted structures blurry textures. To handle this problem, we propose novel method based on wavelet transform attention model...

In order to show the realistic 3D mesh in geometry image-based compression, addition coding image, normal-map image is usually required code. But are difficult compress because it captures more details of original mesh, and has less spatial correlation between pixels than image. This paper proposes a novel framework solve this problem, we effectively predict based on also utilize strong among three components improve predicting accuracy. only need code residual which generated from its...