Few-shot semantic segmentation aims to segment novel-class objects in a query image with only few annotated examples support images. Although progress has been made recently by combining prototype-based metric learning, existing methods still face two main challenges. First, various intra-class between the and images or semantically similar inter-class can seriously harm performance due their poor feature representations. Second, latent novel classes are treated as background most methods,...

Text sentiment analysis is an important task in natural language processing and has always been a hot research topic. However, low-resource regions such as South Asia, where languages like Bengali are widely used, the interest relatively low compared to high-resource due limited computational resources, flexible word order, high inflectional nature of language. With development quantum technology, machine learning models leverage superposition property qubits enhance model expressiveness...

The ability to capture long-distance dependencies is critical for improving the prediction accuracy of spatiotemporal models. Traditional ConvLSTM models face inherent limitations in this regard, along with challenge information decay, which negatively impacts performance. To address these issues, paper proposes a QSA-QConvLSTM model, integrates quantum convolution circuits and self-attention mechanisms. mechanism maps query, key, value vectors using variational circuits, effectively...

In this work, we present a supervised object segmentation algorithm for unconstrained video. Instead of arbitrarily picking few frames manual labeling, as in many existing methods, the proposed method selects more reasonable manner, called supervision optimization. For this, formulate principled objective function by inferring propagation error from appearance and motion clues. After construct multilevel model, which consists low-level high-level features. On low level, image pixels are used...

Deep learning has been widely used in remote sensing image segmentation, while a lack of training data remains significant issue. The few-shot segmentation images refers to the segmenting novel classes with few annotated samples. Although method based on meta-learning can get rid dependence large training, generalization ability model is still low. This work presents self-supervised background learner boost capacity for unseen categories handle this challenge. methodology paper divided into...

Nowcasting has emerged as a critical foundation for services including heavy rain alerts and public transportation management. Although widely used short-term forecasting, models such TrajGRU PredRNN exhibit limitations in predicting low-intensity rainfall low temporal resolution, resulting suboptimal performance during infrequent events. To tackle these challenges, we introduce spatio-temporal sequence generative adversarial network model precipitation forecasting based on radar data. By...

In the era of noisy intermediate-scale quantum (NISQ) computing, synergistic collaboration between and classical computing models has emerged as a promising solution for tackling complex computational challenges. Long short-term memory (LSTM), popular network modeling sequential data, been widely acknowledged its effectiveness. However, with increasing demand data spatial feature extraction, training cost LSTM exhibits exponential growth. this study, we propose convolutional long (QConvLSTM)...

Few-shot semantic segmentation uses a few annotated data of specific class in the support set to segment target same query set. Most existing approaches fail perform well when there are significant intra-class variances. This paper alleviates problem by concentrating on mining image and using as supplementary information. First, it proposes Query Prototype Generation Module generate foreground prototype from features. Specifically, we use both prototype-level pixel-level similarity matching...

In this study, we propose an effective and efficient algorithm for unconstrained video object segmentation, which is achieved in a Markov random field (MRF). the MRF graph, each node modeled as superpixel labeled either foreground or background during segmentation process. The unary potential computed by learning transductive SVM classifier under supervision few frames. pairwise used spatial-temporal smoothness. addition, high-order based on multinomial event model employed to enhance...

In this research, we present the Spatial-Aware Transformer (SAT), an enhanced implementation of Swin module, purposed to augment global modeling capabilities existing transformer segmentation mechanisms within remote sensing. The current landscape techniques is encumbered by inability effectively model dependencies, a deficiency that especially pronounced in context occluded objects. Our innovative solution embeds spatial information into block, facilitating creation pixel-level...

This paper proposes a Robust and Efficient Memory Network, referred to as REMN, for studying semi-supervised video object segmentation (VOS). Memory-based methods have recently achieved outstanding VOS performance by performing non-local pixel-wise matching between the query memory. However, these two limitations. 1) Non-local could cause distractor objects in background be incorrectly segmented. 2) features with high temporal redundancy consume significant computing resources. For...

Recently, memory-based methods have exhibited remarkable performance in Video Object Segmentation (VOS) by employing non-local pixel-wise matching between the query and memory. Nevertheless, these suffer from two limitations: 1) Non-local can result incorrect segmentation of background distractor objects, 2) memory features with substantial temporal redundancy consume significant computing resources reduce inference speed. To address limitations, we first propose a local attention mechanism...

The Internet of Things has been developing fast since it connects various devices for human beings to use, monitor or configure these devices, which brings great convenience. Generally, the big data generated by sensors IoT applications needs be collected, stored and processed realize services such applications. For execution task from applications, may need a variety datasets tasks access process. However, is still challenge placement method when taking into account resource utilization...

Security and privacy issues have attracted the attention of researchers in field IoT as information processing scale grows sensor networks. Quantum computing, theoretically known an absolutely secure way to store transmit well a speed-up accelerate local or distributed classical algorithms that are hard solve with polynomial complexity computation communication. In this paper, we focus on phase estimation method is crucial realization general multi-party computing model, which able be...