Coarse Features Research Articles

Frequency-Aware Feature Fusion for Dense Image Prediction.

Dense image prediction tasks demand features with strong category information and precise spatial boundary details at high resolution. To achieve this, modern hierarchical models often utilize feature fusion, directly adding upsampled coarse features from deep layers and high-resolution features from lower levels. In this paper, we observe rapid variations in fused feature values within objects, resulting in intra-category inconsistency due to disturbed high-frequency features. Additionally, blurred boundaries in fused features lack accurate high frequency, leading to boundary displacement. Building upon these observations, we propose Frequency-Aware Feature Fusion (FreqFusion), integrating an Adaptive Low-Pass Filter (ALPF) generator, an offset generator, and an Adaptive High-Pass Filter (AHPF) generator. The ALPF generator predicts spatially-variant low-pass filters to attenuate high-frequency components within objects, reducing intra-class inconsistency during upsampling. The offset generator refines large inconsistent features and thin boundaries by replacing inconsistent features with more consistent ones through resampling, while the AHPF generator enhances high-frequency detailed boundary information lost during downsampling. Comprehensive visualization and quantitative analysis demonstrate that FreqFusion effectively improves feature consistency and sharpens object boundaries. Extensive experiments across various dense prediction tasks confirm its effectiveness.

WTDUN: Wavelet Tree-Structured Sampling and Deep Unfolding Network for Image Compressed Sensing

Deep unfolding networks have gained increasing attention in the field of compressed sensing (CS) owing to their theoretical interpretability and superior reconstruction performance. However, most existing deep unfolding methods often face the following issues: 1) they learn directly from single-channel images, leading to a simple feature representation that does not fully capture complex features; and 2) they treat various image components uniformly, ignoring the characteristics of different components. To address these issues, we propose a novel wavelet-domain deep unfolding framework named WTDUN, which operates directly on the multi-scale wavelet subbands. Our method utilizes the intrinsic sparsity and multi-scale structure of wavelet coefficients to achieve a tree-structured sampling and reconstruction, effectively capturing and highlighting the most important features within images. Specifically, the design of tree-structured reconstruction aims to capture the inter-dependencies among the multi-scale subbands, enabling the identification of both fine and coarse features, which can lead to a marked improvement in reconstruction quality. Furthermore, a wavelet domain adaptive sampling method is proposed to greatly improve the sampling capability, which is realized by assigning measurements to each wavelet subband based on its importance. Unlike pure deep learning methods that treat all components uniformly, our method introduces a targeted focus on important subbands, considering their energy and sparsity. This targeted strategy lets us capture key information more efficiently while discarding less important information, resulting in a more effective and detailed reconstruction. Extensive experimental results on various datasets validate the superior performance of our proposed method.

12405 A Severe Case Of Virilization Of Premenopausal Woman Secondary To A Rare Testosterone Producing Sertoli Cell Tumor

Abstract Disclosure: S. Waheed: None. N.J. Vernetti: None. S. Nakhle: None. Introduction: We present a case of a premenopausal woman with elevated testosterone who was diagnosed with an aggressive Sertoli cell tumor of the ovary. Case Description: A 49-year-old multiparous female presented to the clinic with hirsutism and markedly elevated testosterone levels. Her symptoms started a year before the presentation with diffuse alopecia, acne, and increased muscle mass. Last menstrual periods were 3 years, prior to the presentation.Denies any exposure to testosterone herself or indirectly through family. On physical exam, blood pressure 170/122 coarse features, oily skin, and muscular build. She had terminal hair loss and increased hair growth on face and arms. Laboratory data revealed testosterone levels of 9280 ng/dl (reference range premenopausal 18-55 ng/dl), free testosterone > 50 ng/dl (premenopausal0-4.2 ng/dl), LH <0.3 mIU/mL (2.4-12.6 mIU/mL), FSH <0.3 mIU/mL (3.6-12.5 mIU/mL), 17-OHP 2270 ng/dl (15-70 ng/dl), DHEA-S 71 ug/dl (<229 ug/dl), LDL 128 mg/dl, Creatinine 1.29 and hematocrit 56% (34-44.6%). CT abdomen/pelvis showed 10 cm ovarian mass. She underwent Total Abdominal Hysterectomy-Bilateral Salpingo Oophorectomy (TAH-BSO). After the surgery, her testosterone levels normalized and features of hyperandrogenism resolved. Pathology report revealed sex cord stromal tumor Not otherwise specified (NOS) with cellular atypia and possible aggressive behavior. Discussion: Androgen-producing tumors in women are rare neoplasms that can cause secondary virilizing characteristics. Of patients presenting with symptoms of hyperandrogenism, these tumors are found in ∼0.2% of cases. Androgen-producing tumors can arise from the ovary or the adrenal gland. Those arising from the ovary are rare, accounting for <5% of all ovarian tumors. Few cases of steroid cell tumors, not otherwise specified, have been reported in the literature.Testosterone elevation of more than 2 times upper normal limit increases the suspicion for cancer. In our patient, testosterone was 180 times above the normal range for women and 15 times above the normal range for men. With rapid virilization and significant elevation in testosterone, an androgen producing ovarian tumors should be considered. Presentation: 6/2/2024

Hierarchical Keypoints Feature Alignment for Domain Adaptive Pose Estimation

Unsupervised Domain Adaptation (UDA) is a popular research topic in computer vision. One of the most common strategies in this field is to reduce domain shifts through global feature alignment. Unfortunately, we observe its failure on adaptive human pose estimation. From our analysis, we find out two major reasons: the extreme imbalance between keypoints and non-keypoints regions and the high diversity of human skeleton structures. To address these problems, we propose a simple yet effective approach named PoseDA. Our idea is to let the alignment focus on the feature of keypoint regions and enforce the model to learn domain-invariant representations for keypoint prediction. The key component of our PoseDA is a Hierarchical Feature Selection of Keypoints Regions (HFS) module, which consists of Coarse Feature Selection of Keypoints Regions (CFS) and Reliable Feature Selection of Keypoints Regions (RFS). By using HFS, we obtain reliable and compact keypoints features, allowing our model to achieve more effective feature alignment. Under three UDA scenarios, i.e. JTA→MPII, SURREAL→LSP, SURREAL→Human3.6, our PoseDA establishes new state-of-the-art performance. In particular, our PoseDA outperforms the previous best UDA methods by over 7% w.r.t. PCKh on JTA→MPII.

ScDTL: enhancing single-cell RNA-seq imputation through deep transfer learning with bulk cell information.

The increasing single-cell RNA sequencing (scRNA-seq) data enable researchers to explore cellular heterogeneity and gene expression profiles, offering a high-resolution view of the transcriptome at the single-cell level. However, the dropout events, which are often present in scRNA-seq data, remaining challenges for downstream analysis. Although a number of studies have been developed to recover single-cell expression profiles, their performance may be hindered due to not fully exploring the inherent relations between genes. To address the issue, we propose scDTL, a deep transfer learning based approach for scRNA-seq data imputation by harnessing the bulk RNA-sequencing information. We firstly employ a denoising autoencoder trained on bulk RNA-seq data as the initial imputation model, and then leverage a domain adaptation framework that transfers the knowledge learned by the bulk imputation model to scRNA-seq learning task. In addition, scDTL employs a parallel operation with a 1D U-Net denoising model to provide gene representations of varying granularity, capturing both coarse and fine features of the scRNA-seq data. Finally, we utilize a cross-channel attention mechanism to fuse the features learned from the transferred bulk imputation model and U-Net model. In the evaluation, we conduct extensive experiments to demonstrate that scDTL could outperform other state-of-the-art methods in the quantitative comparison and downstream analyses.

DSC-Net: Enhancing Blind Road Semantic Segmentation with Visual Sensor Using a Dual-Branch Swin-CNN Architecture.

In modern urban environments, visual sensors are crucial for enhancing the functionality of navigation systems, particularly for devices designed for visually impaired individuals. The high-resolution images captured by these sensors form the basis for understanding the surrounding environment and identifying key landmarks. However, the core challenge in the semantic segmentation of blind roads lies in the effective extraction of global context and edge features. Most existing methods rely on Convolutional Neural Networks (CNNs), whose inherent inductive biases limit their ability to capture global context and accurately detect discontinuous features such as gaps and obstructions in blind roads. To overcome these limitations, we introduce Dual-Branch Swin-CNN Net(DSC-Net), a new method that integrates the global modeling capabilities of the Swin-Transformer with the CNN-based U-Net architecture. This combination allows for the hierarchical extraction of both fine and coarse features. First, the Spatial Blending Module (SBM) mitigates blurring of target information caused by object occlusion to enhance accuracy. The hybrid attention module (HAM), embedded within the Inverted Residual Module (IRM), sharpens the detection of blind road boundaries, while the IRM improves the speed of network processing. In tests on a specialized dataset designed for blind road semantic segmentation in real-world scenarios, our method achieved an impressive mIoU of 97.72%. Additionally, it demonstrated exceptional performance on other public datasets.

High-resolution multicomponent LFM parameter estimation based on deep learning

This paper addresses the complex challenge of parameter estimation in multi-component Linear Frequency Modulation (LFM) signals by introducing an innovative approach to high-resolution Fractional Fourier Transform (FrFT) parameter estimation, facilitated by convolutional neural networks. Initially, it analyzes the issues of peak shifts and the masking of weaker components due to spectral overlap in the FrFT domain of multi-component LFM signals. Convolutional neural networks are then employed to train and achieve high-resolution representations of FrFT parameters. Specifically, convolutional modules with residual structures are utilized to learn coarse features, while a weighted attention mechanism refines independent features across both channel and spatial dimensions. This approach effectively addresses the challenges posed by spectral peak overlap and frequency shifts in multi-component LFM signals, thereby enhancing the quality of high-resolution parameter estimation. Experimental results demonstrate that the proposed method significantly outperforms traditional methods in processing multi-component LFM signals. Moreover, it exhibits robust detection capabilities for both weak and compact components, thereby underscoring its potential applicability in the field of complex signal processing.

CTNet: A data-driven time-frequency technique for wind turbines fault diagnosis under time-varying speeds

Nonstationary fault signals collected from wind turbine planetary gearboxes and bearings often exhibit close-spaced instantaneous frequencies (IFs), or even crossed IFs, bringing challenges for existing time-frequency analysis (TFA) methods. To address the issue, a data-driven TFA technique, termed CTNet is developed. The CTNet is a novel model that combines a fully convolutional auto-encoder network with the convolutional block attention module (CBAM). In the CTNet, the encoder layer is first designed to extract coarse features of the time-frequency representation (TFR) calculated by the general linear Chirplet transform (GLCT); second, the decoder layer is combined to restore and conserve details of the key time-frequency features; third, the skip connections are designed to accelerate training by linking extracted and reconstructed features; finally, the CBAM is introduced to adaptively explore channel and spatial relationships of the TFR, focusing more on close-spaced or crossed frequency features, and effectively reconstruct the TFR. The effectiveness of the CTNet is validated by numerical signals with close-spaced or crossed IFs, and real-world signals of wind turbine planetary gearbox and bearings. Comparison analysis with state-of-the-art TFA methods shows that the CTNet has high time-frequency resolution in characterizing nonstationary signals and a much better ability to detect wind turbine faults.

COLAFormer: Communicating local–global features with linear computational complexity

Local and sparse attention effectively reduce the high computational cost of global self-attention. However, they suffer from non-global dependency and coarse feature capturing, respectively. While some subsequent models employ effective interaction techniques for better classification performance, we observe that the computation and memory of these models overgrow as the resolution increases. Consequently, applying them to downstream tasks with large resolutions takes time and effort. In response to this concern, we propose an effective backbone network based on a novel attention mechanism called Concatenating glObal tokens in Local Attention (COLA) with a linear computational complexity. The implementation of COLA is straightforward, as it incorporates global information into the local attention in a concatenating manner. We introduce a learnable condensing feature (LCF) module to capture high-quality global information. LCF possesses the following properties: (1) performing a function similar to clustering, aggregating image patches into a smaller number of tokens based on similarity. (2) a constant number of aggregated tokens regardless of the image size, ensuring that it is a linear complexity operator. Based on COLA, we build COLAFormer, which achieves global dependency and fine-grained feature capturing with linear computational complexity and demonstrates impressive performance across various vision tasks. Specifically, our COLAFormer-S achieves 84.5% classification accuracy, surpassing other advanced models by 0.4% with similar or less resource consumption. Furthermore, our COLAFormer-S can achieve a better object detection performance while consuming only 1/4 of the resources compared to other state-of-the-art models. The code and models will be made publicly available.

Hypergraph clustering based multi-label cross-modal retrieval

Most existing cross-modal retrieval methods face challenges in establishing semantic connections between different modalities due to inherent heterogeneity among them. To establish semantic connections between different modalities and align relevant semantic features across modalities, so as to fully capture important information within the same modality, this paper considers the superiority of hypergraph in representing higher-order relationships, and proposes an image-text retrieval method based on hypergraph clustering. Specifically, we construct hypergraphs to capture feature relationships within image and text modalities, as well as between image and text. This allows us to effectively model complex relationships between features of different modalities and explore the semantic connectivity within and across modalities. To compensate for potential semantic feature loss during the construction of the hypergraph neural network, we design a weight-adaptive coarse and fine-grained feature fusion module for semantic supplementation. Comprehensive experimental results on three common datasets demonstrate the effectiveness of the proposed method.

Unraveling the binding problem in working memory: insights from the hierarchical binding model.

The binding problem is a crucial issue in the study of working memory (WM) and remains a central topic of debate among various WM models. Over the past decade, we have explored feature binding within WM, guided by the Hierarchical Binding Model (HBM). This model suggests that WM binding occurs in two stages: an initial implicit binding involving rapid, coarse feature processing, followed by explicit binding where focused attention refines these features via a reentry process. We found that implicit binding is closely related to the attentional processing of features during the perceptual stage. Basic features that can be rapidly and coarsely processed in parallel through spread attention are involuntarily extracted into WM along with the target features, forming a rough bound representation. For explicit binding, we examined the role of attention in retaining explicit binding in WM, emphasizing the unique role of reentry in the HBM. Our findings indicate that WM binding requires additional object attention through the reentry process. These results demonstrate that both implicit and explicit bindings are integral to WM and that the HBM is effective in elucidating the binding mechanisms within WM.

MS-CFFS: Multistage Coarse and Fine Feature Selecton for Advanced Anomaly Detection in IoT Security Networks

MS-CFFS: Multistage Coarse and Fine Feature Selecton for Advanced Anomaly Detection in IoT Security Networks

Content-based quality evaluation of scientific papers using coarse feature and knowledge entity network

Pre-evaluating scientific paper quality aids in alleviating peer review pressure and fostering scientific advancement. Although prior studies have identified numerous quality-related features, their effectiveness and representativeness of paper content remain to be comprehensively investigated. Addressing this issue, we propose a content-based interpretable method for pre-evaluating the quality of scientific papers. Firstly, we define quality attributes of computer science (CS) papers as integrity, clarity, novelty, and significance, based on peer review criteria from 11 top-tier CS conferences. We formulate the problem as two classification tasks: Accepted/Disputed/Rejected (ADR) and Accepted/Rejected (AR). Subsequently, we construct fine-grained features from metadata and knowledge entity networks, including text structure, readability, references, citations, semantic novelty, and network structure. We empirically evaluate our method using the ICLR paper dataset, achieving optimal performance with the Random Forest model, yielding F1 scores of 0.715 and 0.762 for the two tasks, respectively. Through feature analysis and case studies employing SHAP interpretable methods, we demonstrate that the proposed features enhance the performance of machine learning models in scientific paper quality evaluation, offering interpretable evidence for model decisions.

Optimization of Grayscale Lithography for the Fabrication of Flat Diffractive Infrared Lenses on Silicon Wafers.

Grayscale lithography (GSL) is an alternative approach to the standard binary lithography in MEMS fabrication, enabling the fabrication of complicated, arbitrary 3D structures on a wafer without the need for multiple masks and exposure steps. Despite its advantages, GSL's effectiveness is highly dependent on controlled lab conditions, equipment consistency, and finely tuned photoresist (PR) exposure and etching processes. This works presents a thorough investigation of the challenges of GSL for silicon (Si) wafers and presents a detailed approach on how to minimize fabrication inaccuracies, aiming to replicate the intended design as closely as possible. Utilizing a maskless laser writer, all aspects of the GSL are analyzed, from photoresist exposure parameters to Si etching conditions. A practical application of GSL is demonstrated in the fabrication of 4-μm-deep f#/1 Si Fresnel lenses for long-wave infrared (LWIR) imaging (8-12 μm). The surface topography of a Fresnel lens is a good case to apply GSL, as it has varying shapes and size features that need to be preserved. The final fabricated lens profiles show a good match with the initial design, and demonstrate successful etching of coarse and fine features, and demonstrative images taken with an LWIR camera.

MRI-Based Brain Tumor Classification Using a Dilated Parallel Deep Convolutional Neural Network

Brain tumors are frequently classified with high accuracy using convolutional neural networks (CNNs) to better comprehend the spatial connections among pixels in complex pictures. Due to their tiny receptive fields, the majority of deep convolutional neural network (DCNN)-based techniques overfit and are unable to extract global context information from more significant regions. While dilated convolution retains data resolution at the output layer and increases the receptive field without adding computation, stacking several dilated convolutions has the drawback of producing a grid effect. This research suggests a dilated parallel deep convolutional neural network (PDCNN) architecture that preserves a wide receptive field in order to handle gridding artifacts and extract both coarse and fine features from the images. This article applies multiple preprocessing strategies to the input MRI images used to train the model. By contrasting various dilation rates, the global path uses a low dilation rate (2,1,1), while the local path uses a high dilation rate (4,2,1) for decremental even numbers to tackle gridding artifacts and to extract both coarse and fine features from the two parallel paths. Using three different types of MRI datasets, the suggested dilated PDCNN with the average ensemble method performs best. The accuracy achieved for the multiclass Kaggle dataset-III, Figshare dataset-II, and binary tumor identification dataset-I is 98.35%, 98.13%, and 98.67%, respectively. In comparison to state-of-the-art techniques, the suggested structure improves results by extracting both fine and coarse features, making it efficient.

P2AT: Pyramid pooling axial transformer for real-time semantic segmentation

Recently, Transformer-based models have achieved promising results in various vision tasks, due to their ability to model long-range dependencies. However, transformers are computationally expensive, which limits their applications in real-time tasks such as autonomous driving. In addition, efficient local and global feature selection and fusion are vital for accurate dense prediction, especially driving scene understanding tasks. In this paper, we propose a real-time semantic segmentation architecture named Pyramid Pooling Axial Transformer (P2AT). The proposed P2AT takes a coarse feature from the CNN encoder to produce scale-aware contextual features, which are then combined with the multi-level feature aggregation scheme to produce enhanced contextual features. Specifically, we introduce a pyramid pooling axial transformer to capture intricate spatial and channel dependencies, leading to improved performance on semantic segmentation. Then, we design a Bidirectional Fusion module (BiF) to combine semantic information at different levels. Meanwhile, a Global Context Enhancer (GCE) is introduced to compensate for the inadequacy of concatenating different semantic levels. Finally, a decoder block is proposed to help maintain a larger receptive field. We evaluate P2AT variants on three challenging scene-understanding datasets. In particular, our P2AT variants achieve state-of-art results on the Camvid dataset 80.5%, 81.0%, 81.1% for P2AT-S, P2AT-M, and P2AT-L, respectively. Furthermore, our experiments on Cityscapes and Pascal VOC 2012 have demonstrated the efficiency of the proposed architecture. The source code will be available at https://github.com/mohamedac29/P2AT.

Image manipulation detection and localization using multi-scale contrastive learning

Current image tampering detection methods rely on various forgery footprints, such as JPEG artifacts and edge inconsistencies, and use algorithms related to image segmentation. However, these methods have several issues, including over-fitting, focusing on only a few specific forgery footprints, and emphasizing semantically relevant information while ignoring tampering traces. This paper proposes a model for image manipulation detection and localization based on multi-scale contrast learning(MSCL-Net). The model utilizes the differences in feature distributions between tampered and untampered regions to extract a comprehensive tamper trace. It uses a dual-stream structured encoder that incorporates both RGB raw images and SRM noise features. A Feature Cross-Fusion Module (FCFM) is proposed to fuse features for improving feature representation of tampered information. The decoding process involves the use of an Adaptive Self-Attention Module (ASAM) to filter and aggregate relevant context from coarse feature maps. Additionally, a Supervised Contrastive Learning Module (SCLM) is used to expand the difference between tampered and untampered areas. The loss function for multi-loss fusion comprises classification loss, segmentation loss, and multi-scale supervised contrastive loss. This improves the network's understanding of global differences, reduces false positives, weakens semantic information, and enhances the model's ability to locate tampered regions of varying sizes. Extensive experiments are conducted across multiple datasets, demonstrating that our model is robust against attacks and resilient to false-positive predictions at both the image-level and pixel-level. Furthermore, its overall performance exceeds that of state-of-the-art alternatives in reliably detecting and localizing tampered images.

An improved approach for automated cervical cell segmentation with PointRend

Regular screening for cervical cancer is one of the best tools to reduce cancer incidence. Automated cell segmentation in screening is an essential task because it can present better understanding of the characteristics of cervical cells. The main challenge of cell cytoplasm segmentation is that many boundaries in cell clumps are extremely difficult to be identified. This paper proposes a new convolutional neural network based on Mask RCNN and PointRend module, to segment overlapping cervical cells. The PointRend head concatenates fine grained features and coarse features extracted from different feature maps to fine-tune the candidate boundary pixels of cell cytoplasm, which are crucial for precise cell segmentation. The proposed model achieves a 0.97 DSC (Dice Similarity Coefficient), 0.96 TPRp (Pixelwise True Positive Rate), 0.007 FPRp (Pixelwise False Positive Rate) and 0.006 FNRo (Object False Negative Rate) on dataset from ISBI2014. Specially, the proposed method outperforms state-of-the-art result by about 3%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$3\\%$$\\end{document} on DSC, 1%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$1\\%$$\\end{document} on TPRp and 1.4%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$1.4\\%$$\\end{document} on FNRo respectively. The performance metrics of our model on dataset from ISBI2015 are slight better than the average value of other approaches. Those results indicate that the proposed method could be effective in cytological analysis and then help experts correctly discover cervical cell lesions

Multi-scale spatio-temporal feature adaptive aggregation for video-based Person Re-identification

Video-based person re-identification (Re-ID) aims to retrieve the target person from video sequences captured by a distributed camera system. It remains a challenging task due to the reasons such as occlusion and misalignment in the video. To address above problem, many methods are proposed to exploit multi-scale spatio-temporal features in videos. However, established methods typically assign the equal weights to temporal or spatial features at different scales, which significantly diminishes the distinct roles of each feature. In this paper, we propose a novel Multi-scale Feature Aggregation Network (MFANet) for video-based person Re-ID. Specifically, we propose two flexible modules, Multi-scale Temporal Feature Aggregation (MTFA) and Multi-scale Spatial Feature Aggregation (MSFA). These two modules first extract different scales of temporal (dynamic, static) and spatial (coarse and fine) features, and then adaptively assign weights to each feature according to the video sequence. Both of these lightweight modules can be incorporated with 3D Convolutional Neural Network to build our MFANet. Extensive experiments on four public benchmarks demonstrate that MTFA and MSFA improve the performance of baseline architectures, and our MFANet achieves the best performance compared to other state-of-the-art methods.

Attention-Based Two-Branch Hybrid Fusion Network for Medical Image Segmentation

Accurate segmentation of medical images is vital for disease detection and treatment. Convolutional Neural Networks (CNN) and Transformer models are widely used in medical image segmentation due to their exceptional capabilities in image recognition and segmentation. However, CNNs often lack an understanding of the global context and may lose spatial details of the target, while Transformers struggle with local information processing, leading to reduced geometric detail of the target. To address these issues, this research presents a Global-Local Fusion network model (GLFUnet) based on the U-Net framework and attention mechanisms. The model employs a dual-branch network that utilizes ConvNeXt and Swin Transformer to simultaneously extract multi-level features from pathological images. It enhances ConvNeXt’s local feature extraction with spatial and global attention up-sampling modules, while improving Swin Transformer’s global context dependency with channel attention. The Attention Feature Fusion module and skip connections efficiently merge local detailed and global coarse features from CNN and Transformer branches at various scales. The fused features are then progressively restored to the original image resolution for pixel-level prediction. Comprehensive experiments on datasets of stomach and liver cancer demonstrate GLFUnet’s superior performance and adaptability in medical image segmentation, holding promise for clinical analysis and disease diagnosis.