In this paper, we propose a similarity-aware deep adversarial learning (SADAL) approach for facial age estimation. Instead of making full access to the limited training samples which likely leads bias prediction, our SADAL aims seek batches unobserved hard-negative based on existing samples, typically reinforces discriminativeness learned feature representation ages. Motivated by fact that labels are usually correlated in real-world scenarios, carefully develop function well measure distance...

3D face reconstruction from single-view images plays an important role in the field of biometrics, which is a long-standing challenging problem wild. Traditional 3DMM-based methods directly regressed parameters, probably caused that network learned discriminative informative features insufficiently. In this paper, we propose replay attention and data augmentation (RADAN) for dense alignment reconstruction. Unlike traditional mechanism, our module aims to increase sensitivity by adaptively...

In this paper, we propose a multi-granularity feature interaction and relation reasoning network (MFIRRN) which can recover detail-rich 3D face perform more accurate dense alignment in an unconstrained environment. Traditional 3DMM-based methods directly regress parameters, resulting the lack of fine-grained details reconstruction face. To end, use different branches to capture discriminative features at granularities, especially local medium fine granularities. Meanwhile, finer-grained...

In this paper, we propose an end-to-end reasoning-decision networks (RDN) approach for robust face alignment via policy gradient. Unlike the conventional coarse-to-fine approaches which likely lead to bias prediction due poor initialization, our aims learn a by leveraging raw pixels reason subset of shape candidates, sequentially making plausible decisions remove outliers initialization. To achieve this, formulate as Markov decision process defining agent, typically interacts with trajectory...

In this paper, we propose a multi-attribute regression network (MARN) to investigate the problem of face reconstruction, especially in challenging cases when faces undergo large variations including severe poses, extreme expressions, and partial occlusions unconstrained environments. The traditional 3DMM parametric method does not distinguish learning identity, expression, attitude attributes, resulting lacking geometric details reconstructed face. We learn features during 3D reconstruction...

3D object reconstruction from a single-view image is long-standing challenging problem. Previous work was difficult to accurately reconstruct shapes with complex topology which has rich details at the edges and corners. Moreover, previous works used synthetic data train their network, but domain adaptation problems occurred when tested on real data. In this paper, we propose Dynamic Multi-branch Information Fusion Network (DmifNet) can recover high-fidelity shape of arbitrary 2D image....

Single-view three-dimensional (3D) object reconstruction has always been a long-term challenging task. Objects with complex topologies are hard to accurately reconstruct, which makes existing methods suffer from blurring of shape boundaries between multiple components in the object. Moreover, most them cannot balance learning global geometric structure information and local detail information. In this article, we propose multi-scale edge-guided network (MEGLN) utilize edge guiding better...

Recovering 3D human mesh from videos has recently made significant progress. However, most of the existing methods focus on temporal consistency videos, while ignoring spatial representation in complex scenes, thus failing to recover a reasonable and smooth sequence under extreme illumination chaotic backgrounds. To alleviate this problem, we propose two-stage co-segmentation network based discriminative representationfor recovering body meshes videos. Specifically, first stage segments...

Reconstructing high-quality 3D object shape from single-view images is a long-standing and challenging problem. Most existing methods overly focus on the overall geometry of target while ignoring local detail areas contained within object, which results in less refined reconstructed results. In this paper, we propose Hierarchical Feature Learning Network (HFLNet), aims to learn global topology paying more attention details, especially fine-grained details thus achieving reconstruction with...

The tradition pattern matching algorithm need backtrack and compare repeatedly, so that affects efficiency of algorithm. Knuth others put forward KMP in order to promote the matching. Paralle based on MPI is provided this paper, which can get higher efficiency.

In order to improve the efficiency of searching longest common subsequence (LCS), a method finding LCS(here, length LCS p is much smaller than string two strings m) realized in this paper, which transform problem into solving matrix L (p, m), by theorem process computing each element optimized. Algorithm analysis and experimental results show algorithm better dynamic programming method. And parallel based on OpenMP provided speed sloving large-scale data greatly improved.

In this paper, we propose a similarity-aware deep adversarial learning (SADAL) approach for facial age estimation. Instead of making access to limited training samples which likely leads sub-optima, our SADAL seeks sets unobserved and plausible hard-examples based on existing samples, typically reinforces the discriminativeness learned feature descriptor ages. Motivated by fact that labels are usually correlated in real-world applications, carefully develop function approach, dynamically...

Multi-view stereo based on deep learning is mostly dedicated to improving the accuracy of point clouds, whlile complex scenes, occlusion, and other factors limit their reconstruction completeness, especially in area with drastic changes depth direction. In this paper, we propose a multi-view network edge flow (DEF-MVSNet), using reference image as guide dynamically infer coordinates improve completeness. First, ignore boundaries prediction stage generate better initial inference results....

In this work, we propose Geometry Normal Consistency Loss (GNC) for 3D face reconstruction and dense alignment. The existing methods based on the strong constraints of 3DMM parameter regression only consider reducing error between 68 landmarks, while they rarely geometric contour structure relation face. Instead, take into account discrete landmarks as loss constrain by introducing geometry area normal consistency loss, which naturally defines holistic local detail, select inverted triangle...

Speech-driven 3D facial animation aims to generate realistic and vivid animations from speech. However, the scarcity of labeled data tendency existing methods treat this cross-modal mapping problem as a regression task can result in inadequate learning discriminative features This deficiency often leads excessively smooth movements, particularly lip movements. To address these issues enhance accuracy generation while reducing reliance on data, we propose CLTalk, framework based contrastive...