Model Compression for Deep Neural Networks: A Survey

Computers, Journal Year: 2023, Volume and Issue: 12(3), P. 60 - 60

Published: March 12, 2023

Currently, with the rapid development of deep learning, neural networks (DNNs) have been widely applied in various computer vision tasks. However, pursuit performance, advanced DNN models become more complex, which has led to a large memory footprint and high computation demands. As result, are difficult apply real time. To address these issues, model compression focus research. Furthermore, techniques play an important role deploying on edge devices. This study analyzed methods assist researchers reducing device storage space, speeding up inference, complexity training costs, improving deployment. Hence, this paper summarized state-of-the-art for compression, including pruning, parameter quantization, low-rank decomposition, knowledge distillation, lightweight design. In addition, discusses research challenges directions future work.

Language: Английский

---

Automated Endoscopic Image Classification via Deep Neural Network With Class Imbalance Loss

IEEE Transactions on Instrumentation and Measurement, Journal Year: 2023, Volume and Issue: 72, P. 1 - 11

Published: Jan. 1, 2023

Recently, many computer-aided diagnosis (CAD) methods have been proposed to help physicians automatically classify endoscopic images. However, most existing often result in poor performance, especially for the minority classes, when dataset is imbalanced. In this paper, we propose a new CAD method automated image classification by introducing novel class imbalance (CI) loss classical deep neural network (DNN). Specifically, use DNN extract rich feature representations. Given that majority usually dominates prediction error and influences gradient of network, CI considers both frequency probability ground-truth assign weight each sample helps classes contribute more descending training process than classes. Thanks loss, pays attention hard samples. To verify effectiveness our method, conduct comprehensive experiments binary-class task on collected polyp recognition (22,935 images) multi-class public Hyper-Kvasir (10,662 images). Experimental results show competent imbalanced with good performance.

Language: Английский

---

Dual-Constraint Coarse-to-Fine Network for Camouflaged Object Detection

IEEE Transactions on Circuits and Systems for Video Technology, Journal Year: 2023, Volume and Issue: 34(5), P. 3286 - 3298

Published: Sept. 25, 2023

Camouflaged object detection (COD) is an important yet challenging task, with great application values in industrial defect detection, medical care, etc. The challenges mainly come from the high intrinsic similarities between target objects and background. In this paper, inspired by biological studies that consists of two steps, i.e., search identification, we propose a novel framework, named DCNet, for accurate COD. DCNet explores candidate extra object-related edges through constraints (object area boundary) detects camouflaged coarse-to-fine manner. Specifically, first exploit area-boundary decoder (ABD) to obtain initial region cues boundary simultaneously fusing multi-level features backbone. Then, module (ASM) embedded into each level backbone adaptively coarse regions assistance ABD. After ASM, refinement (ARM) utilized identify fine adjacent-level guidance cues. Through deep supervision strategy, can finally localize precisely. Extensive experiments on three benchmark COD datasets demonstrate our superior 12 state-of-the-art methods. addition, shows promising results COD-related tasks, polyp segmentation.

Language: Английский

---

Perceptual Quality Assessment of Enhanced Colonoscopy Images: A Benchmark Dataset and an Objective Method

IEEE Transactions on Circuits and Systems for Video Technology, Journal Year: 2023, Volume and Issue: 33(10), P. 5549 - 5561

Published: March 22, 2023

In colonoscopy, the captured images are usually with low-quality appearance, such as non-uniform illumination, low contrast, etc., due to specialized imaging environment, which may provide poor visual feedback and bring challenges subsequent disease analysis. Many low-light image enhancement (LIE) algorithms have recently proposed improve perceptual quality. However, how fairly evaluate quality of enhanced colonoscopy (ECIs) generated by different LIE remains a rarely-mentioned challenging problem. this study, we carry out pioneering investigation on assessment ECIs. Firstly, considering lack specific datasets, collect 300 diverse contents during real-world conduct rigorous subjective studies compare performance 8 popular methods, resulting in benchmark dataset (named ECIQAD) for Secondly, view distinctive distortion characteristics ECIs, propose an effective no-reference Enhanced Colonoscopy Image Quality (ECIQ) method automatically ECIs via analysis brightness, colorfulness, naturalness, noise. Extensive experiments ECIQAD demonstrate superiority our ECIQ over 14 mainstream methods.

Language: Английский

---

Dual Region Mutual Enhancement Network for Camouflaged Object Detection

Published: Jan. 1, 2025

Language: Английский

---

SCABNet: A Novel Polyp Segmentation Network With Spatial‐Gradient Attention and Channel Prioritization

International Journal of Imaging Systems and Technology, Journal Year: 2025, Volume and Issue: 35(2)

Published: Feb. 6, 2025

ABSTRACT Current colorectal polyps detection methods often struggle with efficiency and boundary precision, especially when dealing of complex shapes sizes. Traditional techniques may fail to precisely define the boundaries these polyps, leading suboptimal rates. Furthermore, flat small blend into background due their low contrast against mucosal wall, making them even more challenging detect. To address challenges, we introduce SCABNet, a novel deep learning architecture for efficient polyps. SCABNet employs an encoder‐decoder structure three blocks: Feature Enhancement Block (FEB), Channel Prioritization (CPB), Spatial‐Gradient Boundary Attention (SGBAB). The FEB applies dilation spatial attention high‐level features, enhancing discriminative power improving model's ability capture patterns. CPB, alternative traditional channel blocks, assigns prioritization weights diverse feature channels. SGBAB replaces conventional mechanisms solution that focuses on map. It Jacobian‐based approach construct learned convolutions both vertical horizontal components This allows effectively understand changes in map across different locations, which is crucial detecting complex‐shaped These blocks are strategically embedded within network's skip connections, capabilities without imposing excessive computational demands. They exploit enhance features at levels: high, mid, low, thereby ensuring wide range has been trained Kvasir‐SEG CVC‐ClinicDB datasets evaluated multiple datasets, demonstrating superior results. code available on: https://github.com/KhaledELKarazle97/SCABNet .

Language: Английский

---

MSPolypNet: A residual multi-scale semantic approach for polyps segmentation

Computers & Electrical Engineering, Journal Year: 2025, Volume and Issue: 123, P. 110224 - 110224

Published: March 11, 2025

Language: Английский

---

An Attention U-Net-Based Improved Clutter Suppression in GPR Images

IEEE Transactions on Instrumentation and Measurement, Journal Year: 2024, Volume and Issue: 73, P. 1 - 11

Published: Jan. 1, 2024

The existence of strong back-ground clutter often masks the desired target response, and thereby significantly affect ground penetrating radar (GPR) detection. This effect is even more pronounced for rough terrain shallow buried targets. Therefore, it essential to eliminate facilitate In this paper, a deep learning based attention U-Net model proposed removal GPR data. technique integrates channel module (CAM) spatial (SAM) with architecture enhance performance. implicitely learns suppress irrelevant clutters while emphasizing target. effectiveness approach validated on synthetic as well measured data through visual inspection quantitive evaluation.

Language: Английский

---

DBMA-Net: A Dual-Branch Multiattention Network for Polyp Segmentation

IEEE Transactions on Instrumentation and Measurement, Journal Year: 2024, Volume and Issue: 73, P. 1 - 16

Published: Jan. 1, 2024

In the early prevention stage of colorectal cancer, utilization automatic polyp segmentation techniques from colonoscopy images has demonstrated efficacy in mitigating misdiagnosis rate. Nonetheless, accurate is always against with various challenges, including presence inconsistent size and morphological changes within classes, limited inter-class contrast, high levels interference. recent years, much methodologies based on convolutional neural networks (CNNs) have been widely introduced to enhance precision segmentation. However, two significant hurdles persist: (1) These methods frequently suffer an inadequate acquisition contextual features, causing insufficient feature representation. (2) There a deficiency recognizing intricate information, such as precise boundaries. Addressing these issues, this paper introduces novel dual-branch multi-attention network, denoted DBMA-Net. Specifically, proposed DBMA-Net primarily dual-encoding path that combines CNN Transformer-based approaches enrich Additionally, attention-based fusion module (AFM) incorporated between path, aimed at optimizing features by supplementing local information global insights. Subsequently, distinct attention mechanisms are features: enhancement (AEM) multi-view (MAM), acquire stronger features. modules serve finer details while extensively exploring enhancing lesion region, thereby further elevating accuracy. Following above optimization, enhanced maps hierarchically integrated across multiple scales multi-scale integration (MFIM) for reconstruction. This strategy not only curtails loss but also aids restoring resolution. Ultimately, comprehensive experiments, comparative ablation studies datasets, validate superior performance network compared most state-of-the-art (SOTA) models.

Language: Английский

---

Multi-scale and multi-path cascaded convolutional network for semantic segmentation of colorectal polyps

Alexandria Engineering Journal, Journal Year: 2024, Volume and Issue: 105, P. 341 - 359

Published: July 6, 2024

Colorectal polyps are structural abnormalities of the gastrointestinal tract that can potentially become cancerous in some cases. The study introduces a novel framework for colorectal polyp segmentation named Multi-Scale and Multi-Path Cascaded Convolution Network (MMCC-Net), aimed at addressing limitations existing models, such as inadequate spatial dependence representation absence multi-level feature integration during decoding stage by integrating multi-scale multi-path cascaded convolutional techniques enhances aggregation through dual attention modules, skip connections, enhancer. MMCC-Net achieves superior performance identifying areas pixel level. Proposed was tested across six public datasets compared against eight SOTA models to demonstrate its efficiency segmentation. MMCC-Net's shows Dice scores with confidence intervals ranging between (77.08, 77.56) (94.19, 94.71) Mean Intersection over Union (MIoU) from (72.20, 73.00) (89.69, 90.53) on databases. These results highlight model's potential powerful tool accurate efficient segmentation, contributing early detection prevention strategies cancer.

Language: Английский

---