Multi-level Attention Mechanism Research Articles

Improved diabetic retinopathy severity classification using squeeze-and-excitation and sparse light weight multi-level attention u-net with transfer learning from xception.

Diabetic Retinopathy (DR) is a significant cause of vision loss in diabetic patients, making early detection and accurate severity classification essential for effective management and prevention. This study aims to develop an enhanced DR severity classification approach using advanced model architectures and transfer learning to improve diagnostic accuracy and support better patient care. We propose a novel model, Xception Squeeze-and-Excitation Sparse Lightweight Multi-Level Attention U-Net (XceSE_SparseLwMLA-UNet), designed to classify DR severity using fundus images from the Messidor 1 and Messidor 2 datasets. The XceSE_SparseLwMLA-UNet integrates several advanced mechanisms: the Squeeze-and-Excitation (SE) mechanism for adaptive feature recalibration, the Sparse Lightweight Multi-Level Attention (SparseLwMLA) mechanism for effective contextual information integration, and transfer learning from the Xception architecture to enhance feature extraction capabilities. The SE mechanism refines channel-wise feature responses, while SparseLwMLA enhances the model's ability to identify complex DR patterns. Transfer learning utilizes pre-trained weights from Xception to improve generalization across DR severity levels. The proposed XceSE_SparseLwMLA-UNet model demonstrates superior performance in DR severity classification, achieving higher accuracy and improved multi-class F1 scores compared to existing models. The model's color-coded segmentation outputs offer interpretable visual representations, aiding medical professionals in assessing DR severity levels. The XceSE_SparseLwMLA-UNet model shows promise for advancing early DR diagnosis and management by enhancing classification accuracy and providing valuable visual insights. Its integration of advanced architectural features and transfer learning contributes to better patient care and improved visual health outcomes.

Combined multi-level context aggregation and attention mechanism method for photovoltaic panel extraction from high resolution remote sensing images

ABSTRACT In the context of global carbon emission reduction, solar photovoltaic (PV) technology is experiencing rapid development. Using high-resolution remote sensing images to accurately obtain PV information over a large region, including location and size, has the advantages of high statistical efficiency and timely data update for the PV energy management. Due to the intra-class diversity of PV panels and the intricate variability in their deployment environments, existing semantic segmentation methods often have problems such as under-segmentation and mis-segmentation. To alleviate these problems, this paper proposes an improved DeepLabv3+ semantic segmentation network to more accurately extract PV panels from high-resolution remote sensing images. With the aim of alleviating under-segmentation, a multi-level context aggregation module is developed. This module can enhance the model’s ability to learn the characteristics of PV panels and their surrounding environment by aggregating rich contextual information from multi-scale and semantic levels. To alleviate the problem of mis-segmentation, a hybrid attention module is introduced. This module sequentially and adaptively adjusts the weight distribution in both the channel and spatial dimensions, thus enabling the model to focus more on the feature information and spatial positions of PV objects. Experiments conducted on a self-constructed Beijing PV segmentation dataset show that the method in this paper has advantages of completeness and accuracy in extracting PV panels compared to the baseline model and current mainstream semantic segmentation network. In addition, the results of experiments on extracting PV panels in real region show that our model also has good stability and generalization capability.

A bearing fault diagnosis method based on M-SSCNN and M-LR attention mechanism

Bearing fault diagnosis is vital for mechanical maintenance and fault prediction. It ensures equipment safety, extends lifespan, reduces maintenance costs, and improves production efficiency. Nevertheless, it should be acknowledged that some existing diagnostic methods have achieved high accuracy rates in certain scenarios. However, the challenge lies in their limited generalization capabilities, which can lead to reduced accuracy when applied to diverse or unseen conditions. In this study, we proposed a new bearing fault diagnosis method to address the issue of low accuracy caused by the inadequate generalization of models in the process of rolling bearing fault diagnosis. The method is based on a multi-scale sliding convolution neural network and multi-level residual attention mechanism, the model exhibits high accuracy, strong generalization capability, and lightweight structure. Firstly, the time domain signal of the bearing vibration is converted into a two-dimensional time–frequency map, and the image is pixel-segmented using superpixel segmentation techniques. Next, a multi-scale parallel convolution approach is used to extract features to improve the adaptability and robustness of the model to objects of different sizes and scales. Sliding convolution is used instead of pooling to avoid the problem of feature loss caused by maximum pooling and average pooling. A multi-level attention mechanism is then introduced for all stacked channels to focus on the more important and critical information of the module, and residual connections are added to prevent degradation of the network performance. Finally, the proposed method is passed through the fully connected layer for classification using the Softmax classifier. Experimental verification using public datasets and experimental data of our research group shows that the proposed method has better performance than the existing diagnostic methods and diagnostic models. The proposed method offers an advanced and innovative solution in the domain of bearing fault diagnosis.

Multilevel Attention Network with Semi-supervised Domain Adaptation for Drug-Target Prediction

Prediction of drug-target interactions (DTIs) is a crucial step in drug discovery, and deep learning methods have shown great promise on various DTI datasets. However, existing approaches still face several challenges, including limited labeled data, hidden bias issue, and a lack of generalization ability to out-of-domain data. These challenges hinder the model's capacity to learn truly informative interaction features, leading to shortcut learning and inferior predictive performance on novel drug-target pairs. To address these issues, we propose MlanDTI, a semi-supervised domain adaptive multilevel attention network (Mlan) for DTI prediction. We utilize two pre-trained BERT models to acquire bidirectional representations enriched with information from unlabeled data. Then, we introduce a multilevel attention mechanism, enabling the model to learn domain-invariant DTIs at different hierarchical levels. Moreover, we present a simple yet effective semi-supervised pseudo-labeling method to further enhance our model's predictive ability in cross-domain scenarios. Experiments on four datasets show that MlanDTI achieves state-of-the-art performances over other methods under intra-domain settings and outperforms all other approaches under cross-domain settings. The source code is available at https://github.com/CMACH508/MlanDTI.

A Meta-adversarial Framework for Cross-Domain Cold-Start Recommendation

The cold-start problem in recommender systems has been facing a great challenge. Cross-domain recommendation can improve the performance of cold-start user recommendations in the target domain by using the rich information of users in the source domain. In cross-domain cold-start recommendation, users in target domain lack sufficient historical behaviors. Existing meta-learning-based methods depend on the feature distribution of training data and limit the adaptability in new tasks. To address these issues, we propose a meta-adversarial framework for cross-domain cold-start recommendation (MAFCDR) . Specifically, we employ a multi-level feature attention mechanism for independently learning the weights of long-term and short-term features to construct preferences of users in source domain. To migrate user representations, we train a meta-adversarial network that utilizes feature embeddings in the source domain as input and enhances the robustness and stability of the model. Then, the personalized bridge function transfers the user preferences in the source domain to the target domain. We build three cross-domain tasks using Amazon dataset and conduct extensive experiments, which demonstrate the effectiveness of the proposed model in cold-start user recommendation.

Graph-based social relation inference with multi-level conditional attention

Social relation inference intrinsically requires high-level semantic understanding. In order to accurately infer relations of persons in images, one needs not only to understand scenes and objects in images, but also to adaptively attend to important clues. Unlike prior works of classifying social relations using attention on detected objects, we propose a MUlti-level Conditional Attention (MUCA) mechanism for social relation inference, which attends to scenes, objects and human interactions based on each person pair. Then, we develop a transformer-style network to achieve the MUCA mechanism. The novel network named as Graph-based Relation Inference Transformer (i.e., GRIT) consists of two modules, i.e., a Conditional Query Module (CQM) and a Relation Attention Module (RAM). Specifically, we design a graph-based CQM to generate informative relation queries for all person pairs, which fuses local features and global context for each person pair. Moreover, we fully take advantage of transformer-style networks in RAM for multi-level attentions in classifying social relations. To our best knowledge, GRIT is the first for inferring social relations with multi-level conditional attention. GRIT is end-to-end trainable and significantly outperforms existing methods on two benchmark datasets, e.g., with performance improvement of 7.8% on PIPA and 9.6% on PISC.

Modified DeeplabV3+ with multi-level context attention mechanism for colonoscopy polyp segmentation

The development of automated methods for analyzing medical images of colon cancer is one of the main research fields. A colonoscopy is a medical treatment that enables a doctor to look for any abnormalities like polyps, cancer, or inflammatory tissue inside the colon and rectum. It falls under the category of gastrointestinal illnesses, and it claims the lives of almost two million people worldwide. Video endoscopy is an advanced medical imaging approach to diagnose gastrointestinal disorders such as inflammatory bowel, ulcerative colitis, esophagitis, and polyps. Medical video endoscopy generates several images, which must be reviewed by specialists. The difficulty of manual diagnosis has sparked research towards computer-aided techniques that can quickly and reliably diagnose all generated images. The proposed methodology establishes a framework for diagnosing coloscopy diseases. Endoscopists can lower the risk of polyps turning into cancer during colonoscopies by using more accurate computer-assisted polyp detection and segmentation. With the aim of creating a model that can automatically distinguish polyps from images, we presented a modified DeeplabV3+ model in this study to carry out segmentation tasks successfully and efficiently. The framework's encoder uses a pre-trained dilated convolutional residual network for optimal feature map resolution. The robustness of the modified model is tested against state-of-the-art segmentation approaches. In this work, we employed two publicly available datasets, CVC-Clinic DB and Kvasir-SEG, and obtained Dice similarity coefficients of 0.97 and 0.95, respectively. The results show that the improved DeeplabV3+ model improves segmentation efficiency and effectiveness in both software and hardware with only minor changes.

Spatio-temporal multi-level attention crop mapping method using time-series SAR imagery

Accurate crop mapping is of great significance for crop yield forecasting, agricultural productivity development and agricultural management. Thanks to its all-time and all-weather capability, integrating multi-temporal synthetic aperture radar (SAR) for crop mapping has become essential and challenging task in remote sensing. In recent years, deep learning (DL) has demonstrated excellent crop mapping accuracy to interpret crop dynamics. However, existing DL-based methods tend to be incapable of capturing spatial and temporal features at different scales simultaneously, and this often leads to severe mis-classification due to the complex and heterogeneous distribution of crops and diverse phenological patterns. In this paper, we propose a novel spatio-temporal multi-level attention method, named as STMA, for crop mapping using time-series SAR imagery in an end-to-end fashion to increase the capability of crop phenology retrieval. Specifically, the multi-level attention mechanism is designed to aggregate multi-scale spatio-temporal representations on crops via cascaded spatio-temporal self-attention (STSA) and multi-scale cross-attention (MCA) modalities. To ensure a fine extraction of multi-granularity features, a learnable spatial attention position encoding is proposed to adaptively generate the position priors to facilitate multi-level attention learning. Experimental results on Brandenburg Sentinel-1 dataset, public PASTIS-R dataset and South Africa dataset demonstrated that STMA can achieve state-of-the-art performance in crop mapping tasks, with the accuracy of 96.54% in the Brandenburg Sentinel-1 dataset, 86.77% in the PASTIS-R dataset and 83.37% in the South Africa dataset, validating its effectiveness and superiority. Further comparison of spatio-temporal generalization capability reflected its excellent performance in spatio-temporal modeling on different crops and scenarios. This research provides a viable and intelligent spatio-temporal framework for large-area crop mapping using time-series SAR imagery in complex agricultural systems. The Brandenburg Sentinel-1 dataset and the STMA code will be publicly available at https://github.com/hanzhu97702/ISPRS_STMA.

RelNet-MAM: Relation Network with Multilevel Attention Mechanism for Image Captioning

Text present in an image contains rich semantic information which is crucial for the understanding of an image. For example, a signboard having the text “deep water” conveys the danger involved in the image. The current image captioning models do not effectively utilize this useful semantic information due to their limited representation capabilities of scene-text tokens. Our work presents a novel image captioning model called RelNet-MAM, which utilizes a multilevel attention mechanism and relation network. To improve the appearance feature representation, RelNet-MAM uses multilevel attention which consists of spatial attention, channel-wise attention, and semantic attention. For representing the scene-text token effectively, RelNet-MAM uses appearance, FastText, location, and PHOC features for each token. Further, the proposed RelNet-MAM uses the relation network to establish the relationships between the objects and scene-text tokens. Finally, the transformer model together with dynamic pointer networks is used as a decoder in the caption generation process. The proposed RelNet-MAM model outperforms the state-of-the-art models on TextCaps, Flickr30k, and MS COCO datasets. TextCaps requires models to read and reason about the texts in an image for caption generation. MSCOCO and Flickr30k contain diverse images; persons, animals, automobiles, indoor and outdoor scenes. Remarkably, the proposed RelNet-MAM model outperforms the current best model by 2.3% on B-4, 1.8% on METEOR, 2.2% on ROUGE-L, 2.0% on CIDEr-D and 3.0% on SPICE metric scores on TextCaps dataset.

HiGRN: A Hierarchical Graph Recurrent Network for Global Sea Surface Temperature Prediction

Sea surface temperature (SST) is one critical parameter of global climate change, and accurate SST prediction is important to various applications, e.g., weather forecasting, fishing directions, and disaster warnings. The global ocean system is unified and complex, and the SST patterns in different oceanic regions are highly diverse and correlated. However, existing data-driven SST prediction methods mainly consider the local patterns within a certain oceanic region, e.g., El Nino region and the Black sea. It is challenging but necessary to model the global SST correlations rather than that in a specific region to enhance the prediction accuracy of SST. In this work, we proposed a new method called Hierarchical Graph Recurrent Network (HiGRN) to address the issue. First, to learn the dynamic and diverse local SST patterns of specific locations, we design an adaptive node embedding with self-learned parameters to learn various SST patterns. Then we develop a hierarchical cluster generator to aggregate the locations with similar patterns into regional clusters and utilize a graph convolution network to learn the spatial correlations among these clusters. Finally, we introduce a multi-level attention mechanism to fuse the local patterns and regional correlations, and the output is fed into a recurrent network to achieve SST predictions. Extensive experiments on two real-world datasets show that our method largely outperforms the state-of-the-art SST prediction methods. The source code is available at https://github.com/Neoyanghc/HiGRN .

Price-aware enhanced dynamic recommendation based on deep learning

Price is one of the essential elements influencing consumer purchase behavior. Like consumers’ preferences in products, their price preferences also dynamically change over time. However, dynamic price preferences haven’t been fully considered in existed recommendation studies. In this study, we propose a deep learning-based dynamic recommendation model by considering consumers’ dynamic preferences in both product and price. We specially design a review-and-rating-based sequence generator to select products whose prices the consumers are satisfied with to form the new purchase sequence. We also develop a multi-level attention mechanism in the transformer layer to explore the correlations between consumers’ price choices and to combine the price preferences with the product preferences. Experimental results show the proposed model outperforms the state-of-the-art models on some real-world datasets. Our findings can help retailers understand consumers’ price preferences and make informed decisions related to pricing, discounting, and bundle sales strategies.

For cervical cancer diagnosis: Tissue Raman spectroscopy and multi-level feature fusion with SENet attention mechanism

Cervical cancer ranks among the most prevalent forms of gynecological malignancies. Timely identification of cervical lesions and prompt intervention can effectively prevent the development of cervical cancer or enhance patients' chances of survival. In this study, we propose an innovative method based on Raman spectroscopy, i.e., a multi-level SENet attention mechanism feature fusion architecture (MAFA) for rapid diagnosis of cervical cancer and precancerous lesions. The convolution process of this architecture can extract features from shallow to deep layers, and the attention mechanism is added to achieve the fusion of features from different layers. The added attention mechanism can automatically determine the importance of each layer feature channel and assign weight values to that layer according to the importance of each layer to achieve the purpose of focusing the model on certain waveform features and improve the targeting of model learning. We collected Raman spectra of 212 cervical tissues containing cervical cancer and its precancerous lesions.The experimental results show that MAFA can effectively improve the diagnostic accuracy of VGGNet, GoogLeNet and ResNet models in the validation of Raman spectral data of cervical tissue. Among them, ResNet performed the best, with the highest average accuracy, precision, recall and F1-Score of 82.36%, 84.00%, 82.35% and 82.26%, respectively, when no feature fusion was performed. The evaluation metrics improved by 4.91%, 3.97%, 4.97%, and 5.06%, respectively, after using the MAFA; they also improved by 4.16%, 2.90%, 4.17%, and 4.32%, respectively, compared with the model that directly performs feature fusion without using the attention mechanism. Therefore, the MAFA proposed in this study is better than that of the neural network that directly fuses the features of each convolutional layer. The experimental results show that the performance of the MAFA proposed in this paper is significantly higher than that of traditional deep learning algorithms, indicating that the present architecture can effectively improve the diagnostic accuracy of deep learning networks for cervical cancer.

Improving scene text image captioning using transformer-based multilevel attention

Many existing image captioning methods only focus on image objects and their relationships for generating image captions, ignoring the text present in an image. Scene text (ST) contains crucial information to understand an image and facilitating reasoning. The existing methods fail to establish strong correlations between optical character recognition (OCR) tokens, as they have limited OCR representation power. Further, these methods have not efficiently used the positional information of the text. In this work, we have proposed an ST-based image captioning model (Trans-MAtt) based on a multilevel attention mechanism and relation network. We have used relation networks to enhance the connections between ST tokens. We have employed a multi-level attention method, which comprises of spatial, semantic, and appearance attention modules that precisely define the image. To represent context-enriched ST tokens, we use a combination of appearance, location, FastText, and PHOC features. We predict the ST location in the image, which is further integrated with the generated word embeddings for final caption generation. Experiments on the TextCaps dataset demonstrate the effectiveness of the proposed Trans-MAtt model, where it outperforms the current best model by 3.4% on B-4, 2.9% on METEOR, 3.3% on ROUGE-L, 3.1% on CIDEr-D, and 4.1% on SPICE metric scores. Our experiments on the Flickr30k and MSCOCO datasets demonstrated the superiority of our proposed model over existing methods.

PPLC-Net:Neural network-based plant disease identification model supported by weather data augmentation and multi-level attention mechanism

The accurate detection and identification of plant diseases is an essential step in the development of intelligent and modernized agricultural production. This study proposes a deep learning model (PPLC-Net) incorporating dilated convolution, multi-level attention mechanism, and GAP layers. The model uses novel weather data augmentation to expand the sample size to enhance the generalization and robustness of feature extraction. The feature extraction network extends the perceptual field of the convolutional domain using sawtooth dilated convolution with a variable expansion rate, which can effectively address the problem of insufficient spatial information extraction. The lightweight CBAM attention mechanism is located in the middle layer of the feature extraction network. It is used to enhance the information representation of the model. the GAP layer prevents over-fitting of the model by reducing the number and complexity of parameters computed by the network. The validation of the retained test dataset shows that the recognition accuracy and F1 score of the PPLC-Net model are 99.702% and 98.442%, and the number of parameters and FLOPs are 15.486 M and 5.338G, respectively, which can meet the requirements of accurate and fast recognition. In addition, the proposed combined CAM visualization method can fully validate the effectiveness of the proposed model.

Health insurance fraud detection by using an attributed heterogeneous information network with a hierarchical attention mechanism

BackgroundWith the rapid growth of healthcare services, health insurance fraud detection has become an important measure to ensure efficient use of public funds. Traditional fraud detection methods have tended to focus on the attributes of a single visit and have ignored the behavioural relationships of multiple visits by patients.MethodsWe propose a health insurance fraud detection model based on a multilevel attention mechanism that we call MHAMFD. Specifically, we use an attributed heterogeneous information network (AHIN) to model different types of objects and their rich attributes and interactions in a healthcare scenario. MHAMFD selects appropriate neighbour nodes based on the behavioural relationships at different levels of a patient’s visit. We also designed a hierarchical attention mechanism to aggregate complex semantic information from the interweaving of different levels of behavioural relationships of patients. This increases the feature representation of objects and makes the model interpretable by identifying the main factors of fraud.ResultsExperimental results using real datasets showed that MHAMFD detected health insurance fraud with better accuracy than existing methods.ConclusionsExperiment suggests that the behavioral relationships between patients’ multiple visits can also be of great help to detect health care fraud. Subsequent research fraud detection methods can also take into account the different behavioral relationships between patients.

Multi-level Attention-based Domain Disentanglement for BCDR

Cross-domain recommendation aims to exploit heterogeneous information from a data-sufficient domain (source domain) to transfer knowledge to a data-scarce domain (target domain). A majority of existing methods focus on unidirectional transfer that leverages the domain-shared information to facilitate the recommendation of the target domain. Nevertheless, it is more beneficial to improve the recommendation performance of both domains simultaneously via a dual transfer learning schema, which is known as bidirectional cross-domain recommendation (BCDR). Existing BCDR methods have their limitations, since they only perform bidirectional transfer learning based on domain-shared representations while neglecting rich information that is private to each domain. In this article, we argue that users may have domain-biased preferences due to the characteristics of that domain. Namely, the domain-specific preference information also plays a critical role in the recommendation. To effectively leverage the domain-specific information, we propose a M ulti-level A ttention-based D omain D isentanglement framework dubbed MADD for BCDR, which explicitly leverages the attention mechanism to construct personalized preference with both domain-invariant and domain-specific features obtained by disentangling raw user embeddings. Specifically, the domain-invariant feature is exploited by domain-adversarial learning while the domain-specific feature is learned by imposing an orthogonal loss. We then conduct a reconstruction process on disentangled features to ensure semantic-sufficiency. After that, we devise a multi-level attention mechanism for these disentangled features, which determines their contributions to the final personalized user preference embedding by dynamically learning the attention scores of individual features. We train the model in a multi-task learning fashion to benefit both domains. Extensive experiments on real-world datasets demonstrate that our model significantly outperforms state-of-the-art cross-domain recommendation approaches.

Multi-Level Attention Based Coreference Resolution With Gated Recurrent Unit and Convolutional Neural Networks

Aiming at the diversity of the entity mentions in the field of animal husbandry in Tibet, the reference resolution model Att-GRU-CNN based on multi-level attention mechanism is proposed. The model uses GRU network for global semantic feature learning and knowledge memory, and uses CNN network to further extract local high-level semantic features. A word-level attention layer is added on the GRU hidden layer, the feature vector of entity mention does dot product operation with the feature vector of each word, and the result is normalized and used as the weight value distribution of the words, thus, the prior knowledge that the mention is the most important in the context is given to the network; A sentence-level attention layer is added on the CNN convolution layer, the convolutional layer output and entity name library matrix do association operation, and the result is used as the weight value distribution of the sentence, so as to strengthen the relation between the sentence where the mention is located and the entity scientific name. At the same time, the data enhancement technology is used to improve the generalization ability of the model. Finally, the ablation experiment on Tibetan animal husbandry dataset verified the effectiveness of each component of the model and the superposition effect after combination; the performance comparison experiment is carried out on public datasets MUC, B3 and CEAF <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\varphi _{4}$ </tex-math></inline-formula> . The experimental results show that this model has a significant improvement over other models in the accuracy rate, recall rate and F1 score.

A multimodal fusion model with multi-level attention mechanism for depression detection

Depression is a common mental illness that affects the physical and mental health of hundreds of millions of people around the world. Therefore, designing an efficient and robust depression detection model is an urgent research task. In order to fully extract depression features, we systematically analyze audio-visual and text data related to depression, and proposes a multimodal fusion model with multi-level attention mechanism (MFM-Att) for depression detection. The method is mainly divided into two stages: the first stage utilizes two LSTMs and a Bi-LSTM with attention mechanism to learn multi-view audio feature, visual feature and rich text feature, respectively. In the second stage, the output features of the three modalities are sent into the attention fusion network (AttFN) to obtain effective depression information, aiming to make use of the diversity and complementarity between modalities for depression detection. It is worth noting that the multi-level attention mechanism can not only extract valuable depressive features of intra-modality, but also learn the correlations of inter-modality, thereby improving the overall performance of the model by reducing the influence of redundant information. MFM-Att model is evaluated on the DAIC-WOZ dataset, and the result outperforms state-of-the-art models in terms of root mean square error (RMSE).

A Framework for Image Captioning Based on Relation Network and Multilevel Attention Mechanism

A Framework for Image Captioning Based on Relation Network and Multilevel Attention Mechanism

MFNet: Multiclass Few-Shot Segmentation Network With Pixel-Wise Metric Learning

In visual recognition tasks, few-shot learning requires the ability to learn object categories with few support examples. Its re-popularity in light of the deep learning development is mainly in image classification. This work focuses on few-shot semantic segmentation, which is still a largely unexplored field. A few recent advances are often restricted to single-class few-shot segmentation. In this paper, we first present a novel multi-way (class) encoding and decoding architecture which effectively fuses multi-scale query information and multi-class support information into one query-support embedding. Multi-class segmentation is directly decoded upon this embedding. For better feature fusion, a multi-level attention mechanism is proposed within the architecture, which includes the attention for support feature modulation and attention for multi-scale combination. Last, to enhance the embedding space learning, an additional pixel-wise metric learning module is introduced with triplet loss formulated on the pixel-level of the query-support embedding. Extensive experiments on standard benchmarks PASCAL- <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$5^{i}$ </tex-math></inline-formula> and COCO- <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$20^{i}$ </tex-math></inline-formula> show clear benefits of our method over the state of the art in multi-class few-shot segmentation.