Object Detection Tasks Research Articles

Benchmarking Automated Machine Learning (AutoML) Frameworks for Object Detection

Automated Machine Learning (AutoML) is a subdomain of machine learning that seeks to expand the usability of traditional machine learning methods to non-expert users by automating various tasks which normally require manual configuration. Prior benchmarking studies on AutoML systems—whose aim is to compare and evaluate their capabilities—have mostly focused on tabular or structured data. In this study, we evaluate AutoML systems on the task of object detection by curating three commonly used object detection datasets (Open Images V7, Microsoft COCO 2017, and Pascal VOC2012) in order to benchmark three different AutoML frameworks—namely, Google’s Vertex AI, NVIDIA’s TAO, and AutoGluon. We reduced the datasets to only include images with a single object instance in order to understand the effect of class imbalance, as well as dataset and object size. We used the metrics of the average precision (AP) and mean average precision (mAP). Solely in terms of accuracy, our results indicate AutoGluon as the best-performing framework, with a mAP of 0.8901, 0.8972, and 0.8644 for the Pascal VOC2012, COCO 2017, and Open Images V7 datasets, respectively. NVIDIA TAO achieved a mAP of 0.8254, 0.8165, and 0.7754 for those same datasets, while Google’s VertexAI scored 0.855, 0.793, and 0.761. We found the dataset size had an inverse relationship to mAP across all the frameworks, and there was no relationship between class size or imbalance and accuracy. Furthermore, we discuss each framework’s relative benefits and drawbacks from the standpoint of ease of use. This study also points out the issues found as we examined the labels of a subset of each dataset. Labeling errors in the datasets appear to have a substantial negative effect on accuracy that is not resolved by larger datasets. Overall, this study provides a platform for future development and research on this nascent field of machine learning.

Exploring Object Detection and Image Classification Tasks for Niche Use Case in Naturalistic Driving Studies

Exploring Object Detection and Image Classification Tasks for Niche Use Case in Naturalistic Driving Studies

A novel dual-granularity lightweight transformer for vision tasks

Transformer-based networks have revolutionized visual tasks with their continuous innovation, leading to significant progress. However, the widespread adoption of Vision Transformers (ViT) is limited due to their high computational and parameter requirements, making them less feasible for resource-constrained mobile and edge computing devices. Moreover, existing lightweight ViTs exhibit limitations in capturing different granular features, extracting local features efficiently, and incorporating the inductive bias inherent in convolutional neural networks. These limitations somewhat impact the overall performance. To address these limitations, we propose an efficient ViT called Dual-Granularity Former (DGFormer). DGFormer mitigates these limitations by introducing two innovative modules: Dual-Granularity Attention (DG Attention) and Efficient Feed-Forward Network (Efficient FFN). In our experiments, on the image recognition task of ImageNet, DGFormer surpasses lightweight models such as PVTv2-B0 and Swin Transformer by 2.3% in terms of Top1 accuracy. On the object detection task of COCO, under RetinaNet detection framework, DGFormer outperforms PVTv2-B0 and Swin Transformer with increase of 0.5% and 2.4% in average precision (AP), respectively. Similarly, under Mask R-CNN detection framework, DGFormer exhibits improvement of 0.4% and 1.8% in AP compared to PVTv2-B0 and Swin Transformer, respectively. On the semantic segmentation task on the ADE20K, DGFormer achieves a substantial improvement of 2.0% and 2.5% in mean Intersection over Union (mIoU) over PVTv2-B0 and Swin Transformer, respectively. The code is open-source and available at: https://github.com/ISCLab-Bistu/DGFormer.git.

Explaining Bounding Boxes in Deep Object Detectors Using Post Hoc Methods for Autonomous Driving Systems.

Deep learning has rapidly increased in popularity, leading to the development of perception solutions for autonomous driving. The latter field leverages techniques developed for computer vision in other domains for accomplishing perception tasks such as object detection. However, the black-box nature of deep neural models and the complexity of the autonomous driving context motivates the study of explainability in these models that perform perception tasks. Moreover, this work explores explainable AI techniques for the object detection task in the context of autonomous driving. An extensive and detailed comparison is carried out between gradient-based and perturbation-based methods (e.g., D-RISE). Moreover, several experimental setups are used with different backbone architectures and different datasets to observe the influence of these aspects in the explanations. All the techniques explored consist of saliency methods, making their interpretation and evaluation primarily visual. Nevertheless, numerical assessment methods are also used. Overall, D-RISE and guided backpropagation obtain more localized explanations. However, D-RISE highlights more meaningful regions, providing more human-understandable explanations. To the best of our knowledge, this is the first approach to obtaining explanations focusing on the regression of the bounding box coordinates.

DoubleAUG: Single-domain Generalized Object Detector in Urban via Color Perturbation and Dual-style Memory

Object detection in urban scenarios is crucial for autonomous driving in intelligent traffic systems. However, unlike conventional object detection tasks, urban-scene images vary greatly in style. For example, images taken on sunny days differ significantly from those taken on rainy days. Therefore, models trained on sunny-day images may not generalize well to rainy-day images. In this article, we aim to solve the single-domain generalizable object detection task in urban scenarios, meaning that a model trained on images from one weather condition should be able to perform well on images from any other weather conditions. To address this challenge, we propose a novel Double AUGmentation (DoubleAUG) method that includes image- and feature-level augmentation schemes. In the image-level augmentation, we consider the variation in color information across different weather conditions and propose a Color Perturbation (CP) method that randomly exchanges the RGB channels to generate various images. In the feature-level augmentation, we propose to utilize a Dual-Style Memory (DSM) to explore the diverse style information on the entire dataset, further enhancing the model’s generalization capability. Extensive experiments demonstrate that our proposed method outperforms state-of-the-art methods. Furthermore, ablation studies confirm the effectiveness of each module in our proposed method. Moreover, our method is plug-and-play and can be integrated into existing methods to further improve model performance.

Detection algorithm for dense small objects in high altitude image

The high altitude images captured by drones often contain small-sized objects, which pose several challenges including significant scale variations and dense clustering. To tackle these challenges in small object detection, this paper presents EM-YOLO (Efficient Switching Network and Multi-Scale Feature Fusion Reinforced YOLO for Small Object Detection), an enhanced algorithm based on the YOLOv5s architecture. Firstly, to enhance the network's adaptability to varying object feature scales, the EM-YOLO redesigns the feature extraction backbone network by using different dilation rates in convolutions within the backbone. This flexible structure allows for improved extraction of object features, enhancing the network's feature extraction capability. Secondly, additional detection layer for small objects and combines it with the dynamic detection head. This combination addresses the issue of small object information loss due to multiple down-sampling operations on the feature maps. Finally, Focal-EIoU Loss is employed to enhance regression accuracy in EM-YOLO. Extensive experiments were conducted on the VisDrone2019-DET dataset. The results of these experiments demonstrate that the improved algorithm achieved a significant increase in performance. Specifically, the algorithm achieves 43.5% and 25.1% on the mAP0.5 and mAP0.5:0.95 metrics, which are 13.9% and 8.9% higher than YOLOv5s algorithm, respectively. Furthermore, comparative experiments with other mainstream algorithms indicate that EM-YOLO outperforms them in the task of small object detection.

Backdoor Attacks with Wavelet Embedding: Revealing and enhancing the insights of vulnerabilities in visual object detection models on transformers within digital twin systems

Given the pervasive use of deep learning models across various domains, ensuring model security has emerged as a critical concern. This paper examines backdoor attacks, a form of security threat that compromises model output by poisoning the training data. Our investigation specifically addresses backdoor attacks on object detection models, vital for security-sensitive applications like autonomous driving and smart city systems. Consequently, such attacks on object detection models could pose significant risks to human life and property. Consequently, backdoor attacks on object detection could pose serious threats to human life and property. To elucidate this security risk, we propose and experimentally evaluate five backdoor attack methods for object detection models. The key findings are: (1) Unnecessary Object Generation: a globally embedded trigger creating false objects in the target class; (2) Partial Misclassification: a trigger causing specific class misclassification; (3) Global Misclassification: a trigger reclassifying all objects into the target class; (4) Specific Object Vanishing: a trigger causing non-detection of certain objects; (5) Object Position Shifting: a trigger causing bounding box shifts for a specific class. To assess attack effectiveness, we introduced the Attack Success Rate (ASR), which can surpass 1 in object detection tasks, thus providing a more accurate reflection of the attack impact. Experimental outcomes indicate that the ASR values of these varied backdoor attacks frequently approach or surpass 1, demonstrating our method’s capacity to impact multiple objects simultaneously. Additionally, to augment trigger stealth, we introduce Backdoor Attack with Wavelet Embedding (BAWE), which discreetly embeds triggers as image watermarks in training data. This embedding method yields more natural triggers with enhanced stealth. Highly stealthy triggers are less detectable, significantly increasing the likelihood of attack success and efficacy. We have developed a Transformer-based network architecture, diverging from traditional neural network frameworks. Our experiments across various object detection datasets highlight the susceptibility of these models and the high success rate of our approaches. This vulnerability poses significant risks to digital twin systems utilizing object detection technology. Our methodology not only enhances trigger stealth but also suits dense predictive tasks and circumvents current neural network backdoor attack detection methods. The experimental findings expose key challenges in the security of object detection models, particularly when integrated with digital twins, offering new avenues for backdoor attack research and foundational insights for devising defense strategies against these attacks.

Improving RGB-infrared object detection with cascade alignment-guided transformer

The integration of multispectral data in object detection, especially visible and infrared images, has been the subject of considerable attention recently. Complementary information from visible (RGB) and infrared (IR) images can ameliorate the challenges posed by variable lighting conditions, rendering them an invaluable resource in many fields, including RGB-IR object detection, RGB-IR semantic segmentation, and RGB-IR crowd counting. However, existing methods still suffer from weak misalignment and fusion imprecision problems. These two problems present significant challenges for accurate object detection. In this paper, our primary focus is to solve the above problems in RGB-IR object detection tasks. Specifically, we first propose a Translation-Scale-Rotation Alignment (TSRA) module to align two modality features from region proposals. Based on the aligned region features, we introduce a Complementary Fusion Transformer (CFT) module to capture the complementary features. These two modules can be coupled in an unified Region of Interest (RoI) detection head called Cascade Alignment-Guided Transformer (CAGT) to obtain the robust fused features. Finally, based on CAGT, a region feature alignment and fusion detector called CAGTDet is constructed for RGB-IR object detection. Through comprehensive experiments on the aerial DroneVehicle dataset, our method effectively mitigates the impact of these two issues, resulting in robust detection results. Moreover, to evaluate the generalization of our method, we also perform experiments on the nature images sampled from the KAIST multispectral pedestrian dataset. The results show that our method surpasses other state-of-the-art methods.

Combining Synthetic Images and Deep Active Learning: Data-Efficient Training of an Industrial Object Detection Model.

Generating synthetic data is a promising solution to the challenge of limited training data for industrial deep learning applications. However, training on synthetic data and testing on real-world data creates a sim-to-real domain gap. Research has shown that the combination of synthetic and real images leads to better results than those that are generated using only one source of data. In this work, the generation of synthetic training images via physics-based rendering is combined with deep active learning for an industrial object detection task to iteratively improve model performance over time. Our experimental results show that synthetic images improve model performance, especially at the beginning of the model's life cycle with limited training data. Furthermore, our implemented hybrid query strategy selects diverse and informative new training images in each active learning cycle, which outperforms random sampling. In conclusion, this work presents a workflow to train and iteratively improve object detection models with a small number of real-world images, leading to data-efficient and cost-effective computer vision models.

Education robot object detection with a brain-inspired approach integrating Faster R-CNN, YOLOv3, and semi-supervised learning.

The development of education robots has brought tremendous potential and opportunities to the field of education. These intelligent machines can interact with students in classrooms and learning environments, providing personalized educational support. To enable education robots to fulfill their roles, they require accurate object detection capabilities to perceive and understand the surrounding environment of students, identify targets, and interact with them. Object detection in complex environments remains challenging, as classrooms or learning scenarios involve various objects, backgrounds, and lighting conditions. Improving the accuracy and efficiency of object detection is crucial for the development of education robots. This paper introduces the progress of an education robot's object detection based on a brain-inspired heuristic method, which integrates Faster R-CNN, YOLOv3, and semi-supervised learning. By combining the strengths of these three techniques, we can improve the accuracy and efficiency of object detection in education robot systems. In this work, we integrate two popular object detection algorithms: Faster R-CNN and YOLOv3. We conduct a series of experiments on the task of education robot object detection. The experimental results demonstrate that our proposed optimization algorithm significantly outperforms individual algorithms in terms of accuracy and real-time performance. Moreover, through semi-supervised learning, we achieve better performance with fewer labeled samples. This will provide education robots with more accurate perception capabilities, enabling better interaction with students and delivering personalized educational experiences. It will drive the development of the field of education robots, offering innovative and personalized solutions for education.

ABYOLOv4: improved YOLOv4 human object detection based on enhanced multi-scale feature fusion

The purpose of human object detection is to obtain the number of people and their position in images, which is one of the core problems in the field of machine vision. However, the high missing detection rate from small- and medium-sized human bodies due to the large variety of human scale in human object detection tasks still influences the performance of human object detection. To solve the above problem, this paper proposed an improved ASPP_BiFPN_YOLOv4 (ABYOLOv4) method to detect human object detection. In detail, Atrous Spatial Pyramid Pooling (ASPP) module was used to replace the original Spatial Pyramid Pooling module to increase the receptive field level of the network and improve the perception ability of multi-scale targets. Then, the original Path Aggregation Network (PANet) multi-scale fusion module was replaced by the self-built bi-layer bidirectional feature pyramid network (Bi-FPN). Meanwhile, a new feature was imported into the proposed model to reuse the mid- and low-level features, which could enhance the ability of the network to express the characteristics of small- and medium-sized targets. Finally, the standard convolution in Bi-FPN was replaced by depth-separable convolution to make the network achieve the balance of accuracy and the number of parameters. To identify the performance of the proposed ABYOLOv4 model, the human object detection experiment is carried out by using the public data set of VOC2007 and VOC2012, the improved YOLOv4 algorithm is 0.5% higher than the original AP algorithm, and the weight file size of the model is reduced by 45.3 M. The experimental results demonstrated that the proposed ABYOLOv4 network has higher accuracy and lower computational cost for human target detection.

An efficient segmentation model for abnormal chicken droppings recognition based on improved deep dual-resolution network.

The characteristics of chicken droppings are closely linked to their health status. In prior studies, chicken droppings recognition is treated as an object detection task, leading to challenges in labeling and missed detection due to the diverse shapes, overlapping boundaries, and dense distribution of chicken droppings. Additionally, the use of intelligent monitoring equipment equipped with edge devices in farms can significantly reduce manual labor. However, the limited computational power of edge devices presents challenges in deploying real-time segmentation algorithms for field applications. Therefore, this study redefines the task as a segmentation task, with the main objective being the development of a lightweight segmentation model for the automated monitoring of abnormal chicken droppings. A total of 60 Arbor Acres broilers were housed in 5 specific pathogen-free cages for over 3 weeks, and 1650 RGB images of chicken droppings were randomly divided into training and testing sets in an 8:2 ratio to develop and test the model. Firstly, by incorporating the attention mechanism, multi-loss function and auxiliary segmentation head, the segmentation accuracy of the DDRNet was enhanced. Then, by employing the group convolution and an advanced knowledge distillation algorithm, a lightweight segmentation model named DDRNet-s-KD was obtained, which achieved a mean Dice coefficient (mDice) of 79.43% and an inference speed of 86.10 frames per second (FPS), showing a 2.91% and 61.2% increase in mDice and FPS compared to the benchmark model. Furthermore, the DDRNet-s-KD model was quantized from 32-bit floating-point values to 8-bit integers and then converted to TensorRT format. Impressively, the weight size of the quantized model was only 13.7 MB, representing an 82.96% reduction compared to the benchmark model. This makes it well-suited for deployment on the edge device, achieving an inference speed of 137.51 FPS on Jetson Xavier NX. In conclusion, the methods proposed in this study show significant potential in monitoring abnormal chicken droppings and can provide an effective reference for the implementation of other agricultural embedded systems.

High-low level task combination for object detection in foggy weather conditions

For the object detection task in foggy weather conditions, image dehazing network is often used as preprocessing method to get a clear input. However, there is not strictly a strong positive correlation between image dehazing task and object detection task. Moreover, the preprocessing module can increase the inference time of the whole model to a certain extent. To alleviate these problems, we propose a novel High-Low level task combination network (HLNet) based on multitask learning, which can learn both high-level and low-level tasks. Specially, instead of restoring the features to clear pixel-wise feature space like common image dehazing method, we opt to perform a restoration in feature level to mitigate the influence of the Batch Normalization (BN) layer of encoder on dehazing task. HLNet jointly learn dehazing task and detection task in an end-to-end fashion, which ensures that the weather-specific information in latent feature space is suppressed. Moreover, we applied the HLNet framework on three different object detection networks, including RetinaNnet, YOLOv3 and YOLOv5s network, and achieved improvements of 1.7 percent, 2.3 percent, and 1.2 percent in mAP respectively. The experimental results demonstrate the effectiveness and generalization ability of our proposed HLNet framework in real foggy scenarios.

PI-YOLO: dynamic sparse attention and lightweight convolutional based YOLO for vessel detection in pathological images.

Vessel density within tumor tissues strongly correlates with tumor proliferation and serves as a critical marker for tumor grading. Recognition of vessel density by pathologists is subject to a strong inter-rater bias, thus limiting its prognostic value. There are many challenges in the task of object detection in pathological images, including complex image backgrounds, dense distribution of small targets, and insignificant differences between the features of the target to be detected and the image background. To address these problems and thus help physicians quantify blood vessels in pathology images, we propose Pathological Images-YOLO (PI-YOLO), an enhanced detection network based on YOLOv7. PI-YOLO incorporates the BiFormer attention mechanism, enhancing global feature extraction and accelerating processing for regions with subtle differences. Additionally, it introduces the CARAFE upsampling module, which optimizes feature utilization and information retention for small targets. Furthermore, the GSConv module improves the ELAN module, reducing model parameters and enhancing inference speed while preserving detection accuracy. Experimental results show that our proposed PI-YOLO network has higher detection accuracy compared to Faster-RCNN, SSD, RetinaNet, YOLOv5 network, and the latest YOLOv7 network, with a mAP value of 87.48%, which is 2.83% higher than the original model. We also validated the performance of this network on the ICPR 2012 mitotic dataset with an F1 value of 0.8678, outperforming other methods, demonstrating the advantages of our network in the task of target detection in complex pathology images.

Joint-Guided Distillation Binary Neural Network via Dynamic Channel-Wise Diversity Enhancement for Object Detection

Through truncating the weights and activations of a deep neural network, conventional binary quantization imposes limitations on the representation capability of the network parameters, which hence deteriorates the detection performance of the network. In this paper, a joint-guided distillation binary neural network via dynamic channel-wise diversity enhancement for object detection (JDBNet) is proposed to mitigate the gap of quantization errors. Our JDBNet includes a dynamic channel-wise diversity scheme and real-valued joint-guided teacher assistance to enhance the representation capability of the binary neural network in the object detection tasks. In the dynamic diversity scheme, the learning channel-wise bias (LCB) layer supports adjusting the magnitude of the parameters in which the sensitivity of the model parameters to the arbitrary quantization method is reduced, thereby improving the diversity expression ability of the feature parameters. In the joint-guided strategy, the single-precision implicit knowledge from the guiding teacher in the multilevel layer is utilized to supervise and penalize the quantitative model, enhancing the fitting performance of parameters in the binary quantized model. Extensive experiments on the PASCAL VOC, MS COCO, and VisDrone-DET datasets demonstrate that our JDBNet outperforms the state-of-the-art binary object detection networks in terms of mean Average Precision.

Gradient-guided hierarchical feature attack for object detector

Deep neural networks (DNNs) are vulnerable to adversarial attacks, which can cause security risks in computer information systems. Feature disruption attacks, as a typical form of adversarial attack, optimize adversarial examples by disrupting the intermediate features extracted by DNN. The existing feature disruption attacks have limitations when it comes to objects of different scales within resolution features, favoring low-resolution feature maps and low-scoring objects. The imbalance above affects their effectiveness in object detection tasks. Gradient-guided Hierarchical Feature Attack (GHFA) is proposed to solve these problems, which disrupts detector-extracted features using gradient-guided feature weighting. GHFA incorporates receptive field scaling and gradient scaling to balance the focus among feature maps and enhance attack performance. The evaluation of GHFA on 9 object detectors demonstrates its superior transferability compared with the 6 comparative methods, surpassing the second-ranked method by 2.4%. Furthermore, the real-world applicability of GHFA is validated available by testing it on a commercial online object detection platform.

Relation Knowledge Distillation by Auxiliary Learning for Object Detection.

Balancing the trade-off between accuracy and speed for obtaining higher performance without sacrificing the inference time is a challenging topic for object detection task. Knowledge distillation, which serves as a kind of model compression techniques, provides a potential and feasible way to handle above efficiency and effectiveness issue through transferring the dark knowledge from the sophisticated teacher detector to the simple student one. Despite demonstrating promising solutions to make harmonies between accuracy and speed, current knowledge distillation for object detection methods still suffer from two limitations. Firstly, most of the methods are inherited or refereed from the frameworks in image classification task, and deploy an implicit manner by imitating or constraining the features from the intermediate layers or the output predictions between the teacher and student models. While little consideration has been raised to the intrinsic relevance of the classification and localization predictions in object detection task. Besides, these methods fail to investigate the relationship between detection and distillation tasks in knowledge distillation pipeline, and they train the whole network by simply integrating losses from these two different tasks through hand-crafted designation parameters. For addressing the aforementioned issues, we propose a novel Relation Knowledge Distillation by Auxiliary Learning for Object Detection (ReAL) method in this paper. Specifically, we first design a prediction relation distillation module which makes the student model directly mimic the output predictions from the teacher one, and conduct self and mutual relation distillation losses to excavate the relation information between teacher and student models. Moreover, for better devolving into the relationship between different tasks in distillation pipeline, we introduce the auxiliary learning into knowledge distillation for object detection and develop a dynamic weight adaptation strategy. Through regarding detection task as primary task and treating distillation task as auxiliary task in auxiliary learning framework, we dynamically adjust and regularize the corresponding weights of the losses for these tasks during the training process. Experiments on MS COCO dataset are conducted using various detector combinations of teacher and student models and the results show that our proposed ReAL can achieve obvious improvement on different distillation model configurations, while performing favorably against state-of-the-arts.

Super-tokens Auto-encoders for image compression and reconstruction in IoT applications

New telecommunications networks are enabling powerful AI applications for smart cities and transport. These applications require real-time processing of large amounts of media data. Sending data to the cloud for processing is very difficult due to latency and energy constraints. Lossy compression can help, but traditional codecs may not provide enough quality or be efficient enough for resource-constrained devices. This paper proposes a new image compression and processing approach based on variational auto-encoders (VAEs). This VAE-based method aims to efficiently compress images while still allowing for high-quality reconstruction and object detection tasks. The encoder is designed to be lightweight and suitable for devices with limited computing power. The decoder is more complex and uses multi-level vector quantization to reconstruct high-resolution images. This approach allows for a simple encoder on edge devices and a powerful decoder on cloud servers. Key contributions include a low-complexity encoder, a new VAE model based on vector quantization, and a framework for using VAEs in IoT. The first experiments on reconstructed images on CelebA and ImageNet100 datasets show promising results in terms of MS-SSIM, PSNR, MSE and rFID compared to the literature and the ability of our approach to be used in IoT applications. Our approach presents results similar to complex algorithms like compression algorithms BPG in term of trade-off rate-distortion, and hierarchical auto-encoder (HQA) in terms of image reconstruction quality

Perceptual Video Coding for Machines via Satisfied Machine Ratio Modeling.

Video Coding for Machines (VCM) aims to compress visual signals for machine analysis. However, existing methods only consider a few machines, neglecting the majority. Moreover, the machine's perceptual characteristics are not leveraged effectively, resulting in suboptimal compression efficiency. To overcome these limitations, this paper introduces Satisfied Machine Ratio (SMR), a metric that statistically evaluates the perceptual quality of compressed images and videos for machines by aggregating satisfaction scores from them. Each score is derived from machine perceptual differences between original and compressed images. Targeting image classification and object detection tasks, we build two representative machine libraries for SMR annotation and create a large-scale SMR dataset to facilitate SMR studies. We then propose an SMR prediction model based on the correlation between deep feature differences and SMR. Furthermore, we introduce an auxiliary task to increase the prediction accuracy by predicting the SMR difference between two images in different quality. Extensive experiments demonstrate that SMR models significantly improve compression performance for machines and exhibit robust generalizability on unseen machines, codecs, datasets, and frame types. Code is available at https://github.com/ywwynm/SMR.

GMConv: Modulating Effective Receptive Fields for Convolutional Kernels.

In convolutional neural networks (CNNs), the convolutions are conventionally performed using a square kernel with a fixed N × N receptive field (RF). However, what matters most to the network is the effective receptive field (ERF), which indicates the extent to which input pixels contribute to an output pixel. Inspired by the property that ERFs typically exhibit a Gaussian distribution, we propose a Gaussian Mask convolutional kernel (GMConv). Specifically, GMConv utilizes the Gaussian function to generate a concentric symmetry mask that is placed over the kernel to refine the RF. We analyze the RFs of CNN kernels in different CNN layers and evaluate our approach through extensive experiments on image classification and object detection tasks. Over several tasks and standard base models, our approach compares favorably against the standard convolution. For instance, using GMConv for AlexNet and ResNet-50, the top-1 accuracy on ImageNet classification is boosted by 0.98% and 0.85% , respectively.