Multi-scale Aggregation Network Research Articles

LRMSNet: A New Lightweight Detection Algorithm for Multi-Scale SAR Objects

In recent years, deep learning has found widespread application in SAR image object detection. However, when detecting multi-scale targets against complex backgrounds, these models often struggle to strike a balance between accuracy and speed. Furthermore, there is a continuous need to enhance the performance of current models. Hence, this paper proposes LRMSNet, a new multi-scale target detection model designed specifically for SAR images in complex backgrounds. Firstly, the paper introduces an attention module designed to enhance contextual information aggregation and capture global features, which is integrated into a backbone network with an expanded receptive field for improving SAR image feature extraction. Secondly, this paper develops an information aggregation module to effectively fuse different feature layers of the backbone network. Lastly, to better integrate feature information at various levels, this paper designs a multi-scale aggregation network. We validate the effectiveness of our method on three different SAR object detection datasets (MSAR-1.0, SSDD, and HRSID). Experimental results demonstrate that LRMSNet achieves outstanding performance with a mean average accuracy (mAP) of 95.2%, 98.9%, and 93.3% on the MSAR-1.0, SSDD, and HRSID datasets, respectively, with only 3.46 M parameters and 12.6 G floating-point operation cost (FLOPs). When compared with existing SAR object detection models on the MSAR-1.0 dataset, LRMSNet achieves state-of-the-art (SOTA) performance, showcasing its superiority in addressing SAR detection challenges in large-scale complex environments and across various object scales.

CGMA-Net: Cross-Level Guidance and Multi-Scale Aggregation Network for Polyp Segmentation.

Colonoscopy is considered the best prevention and control method for colorectal cancer, which suffers extremely high rates of mortality and morbidity. Automated polyp segmentation of colonoscopy images is of great importance since manual polyp segmentation requires a considerable time of experienced specialists. However, due to the high similarity between polyps and mucosa, accompanied by the complex morphological features of colonic polyps, the performance of automatic polyp segmentation is still unsatisfactory. Accordingly, we propose a network, namely Cross-level Guidance and Multi-scale Aggregation (CGMA-Net), to earn a performance promotion. Specifically, three modules, including Cross-level Feature Guidance (CFG), Multi-scale Aggregation Decoder (MAD), and Details Refinement (DR), are individually proposed and synergistically assembled. With CFG, we generate spatial attention maps from the higher-level features and then multiply them with the lower-level features, highlighting the region of interest and suppressing the background information. In MAD, we parallelly use multiple dilated convolutions of different sizes to capture long-range dependencies between features. For DR, an asynchronous convolution is used along with the attention mechanism to enhance both the local details and the global information. The proposed CGMA-Net is evaluated on two benchmark datasets, i.e., CVC-ClinicDB and Kvasir-SEG, whose results demonstrate that our method not only presents state-of-the-art performance but also holds relatively fewer parameters. Concretely, we achieve the Dice Similarity Coefficient (DSC) of 91.85% and 95.73% on Kvasir-SEG and CVC-ClinicDB, respectively. The assessment of model generalization is also conducted, resulting in DSC scores of 86.25% and 86.97% on the two datasets respectively.

Hierarchical attention-guided multiscale aggregation network for infrared small target detection

All man-made flying objects in the sky, ships in the ocean can be regarded as small infrared targets, and the method of tracking them has been received widespread attention in recent years. In search of a further efficient method for infrared small target recognition, we propose a hierarchical attention-guided multiscale aggregation network (HAMANet) in this thesis. The proposed HAMANet mainly consists of a compound guide multilayer perceptron (CG-MLP) block embedded in the backbone net, a spatial-interactive attention module (SiAM), a pixel-interactive attention module (PiAM) and a contextual fusion module (CFM). The CG-MLP marked the width-axis, height-axis, and channel-axis, which can result in a better segmentation effect while reducing computational complexity. SiAM improves global semantic information exchange by increasing the connections between different channels, while PiAM changes the extraction of local key information features by enhancing information exchange at the pixel level. CFM fuses low-level positional information and high-level channel information of the target through coding to improve network stability and target feature utilization. Compared with other state-of-the-art methods on public infrared small target datasets, the results show that our proposed HAMANet has high detection accuracy and a low false-alarm rate.

LDD-Net: Lightweight printed circuit board defect detection network fusing multi-scale features

The current printed circuit board (PCB) defect detection model is difficult to balance accuracy and computational cost and cannot satisfy the requirements of practical applications. In this regard, a lightweight PCB image defect detection network (LDD-Net) with high accuracy is proposed in this study. First, LDD-Net designs a novel lightweight feature extraction network (LFEN), which adds convolutional layers for extracting large-scale feature maps and uses an efficient downsampling method to extract richer defect features. LFEN not only greatly reduces resource requirements, but also improves detection accuracy. Second, LDD-Net devises multi-scale aggregation network (MAN) to facilitate information sharing among feature maps at different scales. In addition, MAN incorporates an efficient attention module to suppress similar background interference. Finally, LDD-Net uses the lightweight decoupling head to detect three groups of shallow feature maps with rich information of the small object. Experimental results on the PCB defect dataset ,and VisDrone public dataset show that the mean average precision (mAP), recall, and precision of LDD-Net outperform state-of-the-art lightweight object detection models. The mAP, computation, number of parameters, and detection speed of LDD-Net are 95.90%, 21.48 G, 5.13 M, and 78.13 frames per second, respectively. LDD-Net effectively balances accuracy, computational cost, and detection speed for applications in edge systems or resource-constrained embedded devices.

Multiscale aggregation network via smooth inverse map for crowd counting

Multiscale aggregation network via smooth inverse map for crowd counting

Multi-scale aggregation networks with flexible receptive fields for melanoma segmentation

Skin lesions Segmentation in dermatoscopic images is an important step in automatic diagnosis of melanoma. Due to various of shape and size in skin lesions, this is still a challenging task. In this paper, a multi-scale aggregation network with flexible receptive fields for melanoma segmentation is proposed. We propose the channel-attention dilated convolution module (CDM) to take full of context information. CDM can flexibly adjust the receptive field to capture multi-scale information in response to lesions of various shapes. In addition, we develop aggregation interaction modules to integrate features of adjacent layers of the encoder, which can reduce differences in input features of skip connection and suppress the noise from redundant information. Sub-pixel convolution is adopted as up-sampling operation for improving the fine granularity of detail features. The proposed model is trained on the ISIC 2018 skin segmentation dataset. Experiments and comparison studies are made, and demonstrate that our method produces better segmentation results than other state-of-the-art models in the evaluation metrics of accuracy (Acc), dice coefficient (Dice), Jaccard index (Jac), sensitivity (Sen) and specificity(Spe). Our method achieved 95.7% Acc, 86.4% Dice coefficient, 81.6% Jac, 91.5% Sen, 96.7% Spe, and could well adapt to the scale changes of lesions.

Coarse-to-Fine Embedded PatchMatch and Multi-Scale Dynamic Aggregation for Reference-Based Super-resolution

Reference-based super-resolution (RefSR) has made significant progress in producing realistic textures using an external reference (Ref) image. However, existing RefSR methods obtain high-quality correspondence matchings consuming quadratic computation resources with respect to the input size, limiting its application. Moreover, these approaches usually suffer from scale misalignments between the low-resolution (LR) image and Ref image. In this paper, we propose an Accelerated Multi-Scale Aggregation network (AMSA) for Reference-based Super-Resolution, including Coarse-to-Fine Embedded PatchMatch (CFE-PatchMatch) and Multi-Scale Dynamic Aggregation (MSDA) module. To improve matching efficiency, we design a novel Embedded PatchMacth scheme with random samples propagation, which involves end-to-end training with asymptotic linear computational cost to the input size. To further reduce computational cost and speed up convergence, we apply the coarse-to-fine strategy on Embedded PatchMacth constituting CFE-PatchMatch. To fully leverage reference information across multiple scales and enhance robustness to scale misalignment, we develop the MSDA module consisting of Dynamic Aggregation and Multi-Scale Aggregation. The Dynamic Aggregation corrects minor scale misalignment by dynamically aggregating features, and the Multi-Scale Aggregation brings robustness to large scale misalignment by fusing multi-scale information. Experimental results show that the proposed AMSA achieves superior performance over state-of-the-art approaches on both quantitative and qualitative evaluations.

MSANet: Multiscale Aggregation Network Integrating Spatial and Channel Information for Lung Nodule Detection.

Improving the detection accuracy of pulmonary nodules plays an important role in the diagnosis and early treatment of lung cancer. In this paper, a multiscale aggregation network (MSANet), which integrates spatial and channel information, is proposed for 3D pulmonary nodule detection. MSANet is designed to improve the network's ability to extract information and realize multiscale information fusion. First, multiscale aggregation interaction strategies are used to extract multilevel features and avoid feature fusion interference caused by large resolution differences. These strategies can effectively integrate the contextual information of adjacent resolutions and help to detect different sized nodules. Second, the feature extraction module is designed for efficient channel attention and self-calibrated convolutions (ECA-SC) to enhance the interchannel and local spatial information. ECA-SC also recalibrates the features in the feature extraction process, which can realize adaptive learning of feature weights and enhance the information extraction ability of features. Third, the distribution ranking (DR) loss is introduced as the classification loss function to solve the problem of imbalanced data between positive and negative samples. The proposed MSANet is comprehensively compared with other pulmonary nodule detection networks on the LUNA16 dataset, and a CPM score of 0.920 is obtained. The results show that the sensitivity for detecting pulmonary nodules is improved and that the average number of false-positives is effectively reduced. The proposed method has advantages in pulmonary nodule detection and can effectively assist radiologists in pulmonary nodule detection.

Context-Aware Multi-Scale Aggregation Network for Congested Crowd Counting.

In this paper, we propose a context-aware multi-scale aggregation network named CMSNet for dense crowd counting, which effectively uses contextual information and multi-scale information to conduct crowd density estimation. To achieve this, a context-aware multi-scale aggregation module (CMSM) is designed. Specifically, CMSM consists of a multi-scale aggregation module (MSAM) and a context-aware module (CAM). The MSAM is used to obtain multi-scale crowd features. The CAM is used to enhance the extracted multi-scale crowd feature with more context information to efficiently recognize crowds. We conduct extensive experiments on three challenging datasets, i.e., ShanghaiTech, UCF_CC_50, and UCF-QNRF, and the results showed that our model yielded compelling performance against the other state-of-the-art methods, which demonstrate the effectiveness of our method for congested crowd counting.

Multiscale Aggregate Networks with Dense Connections for Crowd Counting.

The most advanced method for crowd counting uses a fully convolutional network that extracts image features and then generates a crowd density map. However, this process often encounters multiscale and contextual loss problems. To address these problems, we propose a multiscale aggregation network (MANet) that includes a feature extraction encoder (FEE) and a density map decoder (DMD). The FEE uses a cascaded scale pyramid network to extract multiscale features and obtains contextual features through dense connections. The DMD uses deconvolution and fusion operations to generate features containing detailed information. These features can be further converted into high-quality density maps to accurately calculate the number of people in a crowd. An empirical comparison using four mainstream datasets (ShanghaiTech, WorldExpo'10, UCF_CC_50, and SmartCity) shows that the proposed method is more effective in terms of the mean absolute error and mean squared error. The source code is available at https://github.com/lpfworld/MANet.

Lightweight Image Super-Resolution by Multi-Scale Aggregation

Ultra-high-definition display technology is widely used in broadcasting, but there is a huge contradiction between its ultra-high-resolution content and short storage. Super-Resolution (SR) can effectively alleviate this contradiction. Recently, State-of-the-art image SR approaches leveraging Deep Convolutional Neural Networks (DCNNs) have demonstrated high-quality reconstruction performance. However, most of them suffer from large model parameters, which restricts their practical application. Besides, image SR for large scaling factors (e.g., ×8) is a tricky issue when the parameters diminish. To remedy these issues, we propose the Lightweight Multi-scale Aggregation Network (LMAN) for the image SR, which works well for both small and large scaling factors with limited parameters. Specifically, we propose a Group-wise Multi-scale Block (GMB) in which a group convolution is exploited for extracting and fusing multi-scale features before a channel attention layer to obtain discriminative features. Additionally, we present a novel Hierarchical Spatial Attention (HSA) mechanism to jointly and adaptively fuse local and global hierarchical features for high-resolution image reconstruction. Extensive experiments illustrate that our LMAN achieves superior performance against state-of-the-art methods with similar parameters and in particular for large scaling factors such as 4× and 8×.

Multi-scale aggregation network for temporal action proposals

Temporal action detection is a very challenging and valuable task for video analysis and applications. The detection results, to a great extent, rely on the quality of temporal action proposals. However, temporal actions in videos vary dramatically, e.g. from a fraction of a second to minutes, which causes much difficulties for accurate temporal action proposals. In this paper, we propose a multi-scale aggregation network to overcome those variations for temporal action proposals. Our proposed network generates an actionness score sequence for the input video to automatically perceive the duration of actions, and thus can dynamically generate corresponding lengths of action proposals for them. For more reliable actionness prediction, we propose to adaptively explore the intrinsic short and long dependencies in action by two multi-scale aggregation strategies: unit level multi-scale aggregation and proposal level multi-scale aggregation. We also propose to take the soft labelling to facilitate the actionness prediction for the units near the action boundaries. Extensive experiments on THUMOS14 dataset have demonstrated the effectiveness of our proposed method.