Object Identification and Localization Using Grad-CAM++ with Mask Regional Convolution Neural Network

Inbaraj, X. Alphonse; Villavicencio, Charlyn Nayve; Macrohon, Julio Jerison E.; Jeng, Jyh-Horng; Hsieh, Jer‐Guang

doi:10.3390/electronics10131541

Cited by 15 publications

(6 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Moreover, Grad-CAM methods could be of use in various solutions, not only limited to explainable purposes. In [9], authors use generated CAMs as an intermediate step in a vision object detection model. According to the paper, the solution outperforms all the counterpart methods and can be used in unsupervised environments.…”

Section: Gradient-based Methodsmentioning

confidence: 99%

Adaptation of Grad-CAM Method to Neural Network Architecture for LiDAR Pointcloud Object Detection

Dworak

Baranowski

2022

Energies

View full text Add to dashboard Cite

Explainable Artificial Intelligence (XAI) methods demonstrate internal representation of data hidden within neural network trained weights. That information, presented in a form readable to humans, could be remarkably useful during model development and validation. Among others, gradient-based methods such as Grad-CAM are broadly used in an image processing domain. On the other hand, the autonomous vehicle sensor suite consists of auxiliary devices such as radars and LiDARs, for which existing XAI methods do not apply directly. In this article, we present our adaptation approach to utilize Grad-CAM visualization for LiDAR pointcloud specific object detection architectures used in automotive perception systems. We try to solve data and network architecture compatibility problems and answer the question whether Grad-CAM methods could be used with LiDAR sensor data efficiently. We showcase successful results of our method and all the benefits that come with a Grad-CAM XAI application to a LiDAR sensor in an autonomous driving domain.

show abstract

Section: Gradient-based Methodsmentioning

confidence: 99%

Adaptation of Grad-CAM Method to Neural Network Architecture for LiDAR Pointcloud Object Detection

Dworak

Baranowski

2022

Energies

View full text Add to dashboard Cite

show abstract

“…Taking advantage of this point, many studies have successfully applied this XAI technique. The research uses Grad-CAM++ with the Mask Regional Convolution Neural Network, which has shown a powerful real-time ability to detect and classify existing objects and shapes [24]. The research uses Grad-CAM++ enabled with the squeeze-and-excite network on the CT scan dataset of lung cancer [25].…”

Section: Related Workmentioning

confidence: 99%

Tomato Health Monitoring System: Tomato Classification, Detection, and Counting System Based on YOLOv8 Model With Explainable MobileNet Models Using Grad-CAM++

Quach,

Quoc,

Quynh

et al. 2024

IEEE Access

View full text Add to dashboard Cite

Fruits and vegetables (especially, tomatoes) healthy detection are important tasks for smart agriculture. Several works have been published in tomato detection, however, there is little research on using explainable AI to detect, classify and count tomato fruit status. In this work, we propose a Tomatoes Health Check System by evaluating MobileNet models based on the physiological tomato dataset. The MobileNet model is considered a model specifically designed to operate effectively on mobile devices and limited resources. Therefore, MobileNet models used in applying explainable artificial intelligence (XAI) problems must be considered. Our research conducts experiments to evaluate the accuracy of the MobileNets, MobileNetV2 and MobileNetV3 models based on the evaluation metrics; the highest accuracy of 96.69% belongs to the MobileNetV3 model. To evaluate the reliability of the model used, we use Grad-CAM++ to explain and evaluate reliability based on the Intersection over the Union parameter, resulting in MobileNetV2 being the model with the highest value with 100.00%(δ=0), 100.00% (δ=0.25) and 98.89% (δ=0.5). Our research conducted an evaluation experiment using the Simple online and real-time tracking algorithm to combine the YOLOv8 and MobileNetV2 algorithms to detect, classify and count the number of tomatoes according to physiology using video. Finally, we use the research results to build an application system.

show abstract

“…Class activation map (CAM) is an intuitive method in convolutional neural network (CNN) interpretation, and it was usually generated by the last convolutional layer of a CNN, which can highlight different regions of the target class in the input image. To verify the contribution of CA-MobileNetV3 and CARAFE upsampling module to the count of wheat ears, an object localization method (Grad-CAM++) [32] was used to construct important regions of multiple objects on the image through gradients with convolutional layers for feature visualization. In this study, LWDNet and DM-Count were selected for feature visualization comparison, and the original wheat ear map, the wheat ear heat map of DM-Count and LWDNet, and the wheat ear visualization class heat activation map, were displayed, respectively, as shown in Figure 11.…”

Section: Existing Counting Challenges and Visualization Of Existing T...mentioning

confidence: 99%

Cross-Platform Wheat Ear Counting Model Using Deep Learning for UAV and Ground Systems

et al. 2023

View full text Add to dashboard Cite

Wheat is one of the widely cultivated crops. Accurate and efficient high-throughput ear counting is important for wheat production, yield evaluation, and seed breeding. The traditional wheat ear counting method is inefficient due to the small scope of investigation. Especially in the wheat field scene, the images obtained from different platforms, including ground systems and unmanned aerial vehicles (UAVs), have differences in density, scale, and wheat ear distribution, which makes the wheat ear counting task still face some challenges. To this end, a density map counting network (LWDNet) model was constructed for cross-platform wheat ear statistics. Firstly, CA-MobileNetV3 was constructed by introducing a collaborative attention mechanism (CA) to optimize the lightweight neural network MobileNetV3, which was used as the front end of the feature extraction network, aiming to solve the problem of occlusion and adhesion of wheat ears in the field. Secondly, to enhance the model’s ability to learn the detailed features of wheat ears, the CARAFE upsampling module was introduced in the feature fusion layer to better restore the characteristics of wheat ears and improve the counting accuracy of the model for wheat ears. Finally, density map regression was used to achieve high-density, small-target ear counting, and the model was tested on datasets from different platforms. The results showed that our method can efficiently count wheat ears of different spatial scales, achieving good accuracy while maintaining a competitive number of parameters (2.38 million with a size of 9.24 MB), which will benefit wheat breeding and screening analysis to provide technical support.

show abstract

Object Identification and Localization Using Grad-CAM++ with Mask Regional Convolution Neural Network

Cited by 15 publications

References 21 publications

Adaptation of Grad-CAM Method to Neural Network Architecture for LiDAR Pointcloud Object Detection

Adaptation of Grad-CAM Method to Neural Network Architecture for LiDAR Pointcloud Object Detection

Tomato Health Monitoring System: Tomato Classification, Detection, and Counting System Based on YOLOv8 Model With Explainable MobileNet Models Using Grad-CAM++

Cross-Platform Wheat Ear Counting Model Using Deep Learning for UAV and Ground Systems

Contact Info

Product

Resources

About