Three-dimensional point cloud registration is an important field in computer vision. Recently, due to the increasingly complex scenes and incomplete observations, many partial-overlap methods based on overlap estimation have been proposed. These heavily rely extracted overlapping regions with their performances greatly degraded when region extraction underperforms. To solve this problem, we propose a partial-to-partial network (RORNet) find reliable representations from partially clouds use...

Band selection is an important preprocessing step for hyperspectral image processing. Many valid criteria have been proposed band selection, and these model as a single-objective optimization problem. In this paper, novel multiobjective first built selection. model, two objective functions with conflicting relationship are designed. One function set information entropy to represent the contained in selected subsets, other one number of bands. Then, based on new unsupervised method called...

Feature extraction (FE) is a crucial research area in hyperspectral image (HSI) processing. Recently, due to the powerful ability of deep learning (DL) extract spatial and spectral features, DL-based FE methods have shown great potentials for HSI However, most are supervised, training them suffers from absence labeled samples HSIs severely. The issue supervised limits their application on To address this issue, paper, novel modified generative adversarial network (GAN) proposed train feature...

Due to the lack of label information and intrinsic complexity hyperspectral images (HSIs), unsupervised band selection is always one most challenging tasks in HSI processing. Fuzzy clustering a promising technique for selection, which can partition unlabeled data into groups effectively. However, due limits its optimization process, standard fuzzy sensitive initialization easy be trapped local optimum. To address limits, novel method proposed, combining with particle swarm (PSO). A newly...

Building change detection (BCD) for very-high-spatial-resolution (VHR) remote sensing images is very important and challenging in the field of sensing, as building one most significant valuable man-made ground targets. This article proposes a local–global pyramid network (LGPNet) that combines local feature module (LFPM) global spatial (GSPM) various extraction. The LFPM constructed using convolutional kernel with three different scales, then, features are obtained by adding each...

Extracting 3D information from a single optical image is very attractive. Recently emerging self-supervised methods can learn depth representations without using ground truth maps as training data by transforming the prediction task into an synthesis task. However, existing rely on differentiable bilinear sampler for synthesis, which results in each pixel synthetic being derived only four pixels source and causes map to perceive few image. In addition, when calculating photometric error...

Self-supervised learning is an effective way to solve model collapse for few-shot remote sensing scene classification (FSRSSC). However, most self-supervised contrastive auxiliary tasks perform poorly on the high interclass similarity problem in FSRSSC. Furthermore, it time-consuming and computationally expensive obtain best combination among numerous tasks. In practical applications, we may encounter difficulties data acquisition labeling, while FSRSSC studies only focus former. To...

In recent years, deep learning models, which possess powerful feature extraction abilities, have achieved remarkable success in the classification of hyperspectral images (HSIs). Nevertheless, a common challenge faced by most including few-shot is scarcity valid labeled samples. To address this issue, we propose cross-domain self-taught network (CDSTN) for image classification. The proposed CDSTN merges domain adaptation and semi-supervised strategy to implement learning, utilizes adequate...

The statistical approach to voice conversion typically consists of a feature module followed by vocoder. So far, the studies are mainly focused on spectrum. However, speaker identity is also characterized prosodic features, such as fundamental frequency F0 and energy contour among others. In this paper, we study transformation characteristics both in terms spectrum prosody. We propose two novel techniques that effectively use limited amount source-target training data leverage large general...

Hyperspectral images (HSIs) contain rich spectral signatures that reveal more image details and, thus, enable the detection of less noticeable changes on ground. However, HSI-based change (CD) is susceptible to a large amount irrelevant or noisy and spatial information due massive bands. To address these issues, we propose novel attention network (S²AN) for CD, which capable suppress CD-irrelevant via adaptive mechanisms. S²AN takes as input patch from difference map...

The few-shot scene classification is dedicated to identifying unseen remote sensing classes when only a very small number of labeled samples are available for reference. Most the existing methods based on meta-learning and employ episodic learning training, which lacks consideration utilization data efficiency. In this paper, instead designing sophisticated algorithms, we committed training feature extractor with good generalization performance strong extraction capability. Specifically,...

3D interacting hand pose estimation from a single RGB image is challenging task, due to serious self-occlusion and inter-occlusion towards hands, confusing similar appearance patterns between 2 ill-posed joint position mapping 2D 3D, etc.. To address these, we propose extend A2J-the state-of-the-art depth-based method-to domain under condition. Our key idea equip A2J with strong local-global aware ability well capture hands' local fine details global articulated clues among joints jointly....

Self-supervised contrastive learning (CL) can learn high-quality feature representations that are beneficial to downstream tasks without labeled data. However, most CL methods for image-level tasks. For the fine-grained change detection (FCD) tasks, such as or trend of some specific ground objects, it is usually necessary perform pixel-level discriminative analysis. Therefore, learned by may have limited effects on FCD. To address this problem, we propose a self-supervised global–local...

In this letter, we proposed a novel band selection algorithm for hyperspectral images (HSIs) based on column subset selection. The main idea of the comes from problem in numerical linear algebra. It selects group bands, which maximizes volume selected columns. Since high dimensionality decreases contrast between use Manhattan distance to obtain higher quality. Experimental results real HSIs show that obtains competitively good results, terms classification accuracy, and is robust noisy bands.

Many natural language processing and information retrieval problems can be formalized as the task of semantic matching. Existing work in this area has been largely focused on matching between short texts (e.g., question answering), or a long text ad-hoc retrieval). Semantic long-form documents, which many important applications like news recommendation, related article recommendation document clustering, is relatively less explored needs more research effort. In recent years, self-attention...

One of key challenges skeleton-based action recognition (SAR) tasks is the complex nature human motion patterns. Variations such as performers and viewpoints may impose negative effects to accuracy. In this work, we propose Multi-Localized Sensitive Autoencoder-Attention-LSTM (Multi-LiSAAL) for SAR. The Localized Stochastic Autoencoder (LiSSA) encodes both spatial temporal information, extracts meaningful features from different parts (four limbs a trunk) skeleton. LiSSA trained by...

The content on the web is in a constant state of flux. New entities,issues, and ideas continuously emerge, while semantics existing conversation topics gradually shift. In recent years, pre-trained language models like BERT greatly improved state-of-the-art for large spectrum understanding tasks.Therefore, this paper, we aim to study how these can be adapted better handle evolving content.In our study, first analyze evolution 2013 - 2019 Twitter data, unequivocally confirm that model trained...