Facial Expression Recognition (FER) in the wild is extremely challenging due to occlusions, variant head poses, face deformation and motion blur under unconstrained conditions. Although substantial progresses have been made automatic FER past few decades, previous studies were mainly designed for lab-controlled FER. Real-world poses other issues definitely increase difficulty of on account these information-deficient regions complex backgrounds. Different from pure CNNs based methods, we...

Convolutional neural network (CNN) has achieved great success on image super-resolution (SR). However, most deep CNN-based SR models take massive computations to obtain high performance. Downsampling features for multi-resolution fusion is an efficient and effective way improve the performance of visual recognition. Still, it counter-intuitive in task, which needs project a low-resolution input high-resolution. In this paper, we propose novel Hybrid Pixel-Unshuffled Network (HPUN) by...

Abstract Background We developed a computer-assisted diagnosis model to evaluate the feasibility of automated classification intrapapillary capillary loops (IPCLs) improve detection esophageal squamous cell carcinoma (ESCC). Methods recruited patients who underwent magnifying endoscopy with narrow-band imaging for evaluation suspicious condition. Case images were evaluated establish gold standard IPCL according endoscopic and histological findings. A double-labeling fully convolutional...

Hyperspectral images (HSIs) are often degraded by a mixture of various types noise during the imaging process, including Gaussian noise, impulse and stripes. Such complex could plague subsequent HSIs processing. Generally, most HSI denoising methods formulate sparsity optimization problems with convex norm constraints, which over-penalize large entries vectors, may result in biased solution. In this paper, nonconvex regularized low-rank sparse matrix decomposition (NonRLRS) method is...

Automatic speech recognition (ASR) is the major human–machine interface in many intelligent systems, such as homes, autonomous driving, and servant robots. However, its performance usually significantly deteriorates presence of external noise, leading to limitations application scenes. The audio-visual (AVSR) takes visual information a complementary modality enhance audio effectively, particularly noisy conditions. Recently, transformer-based architectures have been used model video...

Action prediction based on video is an important problem in computer vision field with many applications, such as preventing accidents and criminal activities. It's challenging to predict actions at the early stage because of large variations between observed videos complete ones. Besides, intra-class cause confusions predictors well. In this paper, we propose a mem-LSTM model stage, which memory module introduced record several "hard-to-predict" samples variety observations. Our method uses...

Active learning (AL) and semisupervised (SSL) are both promising solutions to hyperspectral image classification. Given a few initial labeled samples, this work combines AL SSL in novel manner, aiming obtain more manually pseudolabeled samples use them together with the improve classification performance. First, based on comparison of segmentation spectral-spatial results obtained by random walker (RW) extended RW (ERW) algorithms, unlabeled separated into two different sets, i.e., low-...

Image rasterization is a mature technique in computer graphics, while image vectorization, the reverse path of rasterization, remains major challenge. Recent advanced deep learning-based models achieve vectorization and semantic interpolation vector graphs demonstrate better topology generating new figures. However, cannot be easily generalized to out-of-domain testing data. The generated SVGs also contain complex redundant shapes that are not quite convenient for further editing....

Temporal answer grounding in instructional video (TAGV) is a new task naturally derived from temporal sentence general (TSGV). Given an untrimmed and text question, this aims at locating the frame span that can semantically i.e., visual answer. Existing methods tend to solve TAGV problem with span-based predictor, taking information predict start end frames video. However, due weak correlations between semantic features of textual question answer, current using predictor do not work well...

Glioblastoma (GBM) tumor is the most common primary brain malignant tumor. The precise identification of GBM very important for diagnosis and treatment. Hyperspectral imaging a fast, non-contact, accurate safety modern medical detection technology, which expected to be new tool intraoperative diagnosis. In order make full use spectral spatial information hyperspectral images (HSIs) achieve identification, method based on fusion multiple deep models (FMDM) proposed in-vivo human HSI...

Previous methods for dynamic facial expression in the wild are mainly based on Convolutional Neural Networks (CNNs), whose local operations ignore long-range dependencies videos. To solve this problem, we propose spatio-temporal Transformer (STT) to capture discriminative features within each frame and model contextual relationships among frames. Spatio-temporal captured integrated by our unified Transformer. Specifically, given an image sequence consisting of multiple frames as input,...

Additive manufacturing of carbon-fiber-reinforced polymer (CFRP) has been widely used in many fields. However, issues such as inconsistent fiber orientation distribution and void formation during the layer stacking process have hindered further optimization composite material’s performance. This study aimed to address these challenges by conducting a comprehensive investigation into influence carbon content printing parameters on micro-morphology, thermal properties, mechanical properties...

In the context of global warming, sustainability farmland ecosystems is increasingly impacted by multiple disturbances from both natural and human-induced sources. This study constructed a conceptual model indicator system ecosystem resilience (FER) based on disturbance-response processes ecosystems. FER assessment, supported 30 specific indicators, was tested in Ethiopia, one most food-insecure countries world factors impending were discussed obstacle degree values (ODVs). The results...

Grassland is the second largest terrestrial ecosystem and a fundamental land resource for human survival development. Although grassland degradation recognized crucial ecological problem, there no consensus on area, scope, degree of its global trends, making implementation Sustainable Development Goals (SDG) 15.3 achieving degradation-neutral world uncertain. This study quantitatively explored trends from 2000 to 2020 by coupling vegetation growth response climate change. Furthermore,...

Previous methods for dynamic facial expression recognition (DFER) in the wild are mainly based on Convolutional Neural Networks (CNNs), whose local operations ignore long-range dependencies videos. Transformer-based DFER can achieve better performances but result higher FLOPs and computational costs. To solve these problems, local-global spatio-temporal Transformer (LOGO-Former) is proposed to capture discriminative features within each frame model contextual relationships among frames while...

Semisupervised semantic segmentation is an effective way to reduce the expensive manual annotation cost and take advantage of unlabeled data for remote sensing (RS) image interpretation. Recent related research has mainly adopted two strategies: self-training consistency regularization. Self-training tries acquire accurate pseudo-labels explicitly expand train set. However, existing methods cannot accurately identify false pseudo-labels, suffering from their negative impact on model...