Estimation of Road Boundary for Intelligent Vehicles Based on DeepLabV3+ Architecture

Das, Sunanda; Fime, Awal Ahmed; Siddique, Nazmul; Hashem, M. M. A.

doi:10.1109/access.2021.3107353

Cited by 29 publications

(7 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Inspired by [ 44 ] and its work on road boundary estimation, we employed a reduced ResNet50 [ 42 ] network (without the fourth block) for feature extraction. We found that using a backbone model with a small depth yields better results with our dataset.…”

Section: Proposed Methodsmentioning

confidence: 99%

mid-DeepLabv3+: A Novel Approach for Image Semantic Segmentation Applied to African Food Dietary Assessments

Baban A Erep,

Chaari

2023

Sensors

View full text Add to dashboard Cite

Recent decades have witnessed the development of vision-based dietary assessment (VBDA) systems. These systems generally consist of three main stages: food image analysis, portion estimation, and nutrient derivation. The effectiveness of the initial step is highly dependent on the use of accurate segmentation and image recognition models and the availability of high-quality training datasets. Food image segmentation still faces various challenges, and most existing research focuses mainly on Asian and Western food images. For this reason, this study is based on food images from sub-Saharan Africa, which pose their own problems, such as inter-class similarity and dishes with mixed-class food. This work focuses on the first stage of VBDAs, where we introduce two notable contributions. Firstly, we propose mid-DeepLabv3+, an enhanced food image segmentation model based on DeepLabv3+ with a ResNet50 backbone. Our approach involves adding a middle layer in the decoder path and SimAM after each extracted backbone feature layer. Secondly, we present CamerFood10, the first food image dataset specifically designed for sub-Saharan African food segmentation. It includes 10 classes of the most consumed food items in Cameroon. On our dataset, mid-DeepLabv3+ outperforms benchmark convolutional neural network models for semantic image segmentation, with an mIoU (mean Intersection over Union) of 65.20%, representing a +10.74% improvement over DeepLabv3+ with the same backbone.

show abstract

Section: Proposed Methodsmentioning

confidence: 99%

mid-DeepLabv3+: A Novel Approach for Image Semantic Segmentation Applied to African Food Dietary Assessments

Baban A Erep,

Chaari

2023

Sensors

View full text Add to dashboard Cite

show abstract

“…Inspired by [42] and its works on road boundaries estimation, we employ reduced ResNet50 [9] network (without the 4th block) for features extraction. We have found that using a backbone model with a small depth yields better results with our dataset.…”

Section: Backbone and Features Extractionmentioning

confidence: 99%

mid-DeepLabv3+ : A Novel Approach for Image Semantic Segmentation applied to African Food Dietary Assessment

EREP,

CHAARI,

ELE

et al. 2023

Preprint

View full text Add to dashboard Cite

Recent decades have seen the development of vision-based dietary assessment systems (VBDA). These systems generally consist of three main stages: food image analysis, portion estimation and nutrient derivation. The effectiveness of the initial step is highly dependent on the use of accurate segmentation and image recognition models, and the availability of high-quality training datasets. Food image segmentation still faces various challenges, while most existing research has focused mainly on Asian and Western food images. For this reason, this study is based on food images from sub-Saharan Africa, which pose their own problems, such as inter-class similarity and dishes presentation with mixed-class food. This work focuses on the first stage of VBDA, where we introduce two notable contributions. Firstly, we propose mid-DeepLabv3+, an enhanced food image segmentation model based on DeepLabv3+ with a ResNet50 backbone. Our approach involves adding a middle layer in the decoder path and SimAM attention mechanism after each extracted backbone features layers. Secondly, we present CamerFood10, the first food images dataset specifically designed for sub-Saharan African food segmentation. It includes 10 classes of most consumed food items in Cameroon. On our dataset, mid-DeepLabv3+ outperforms existing image segmentation benchmark models with an mIoU (mean Intersection over Union) of 65.20\%, representing more than +10.74\% improvement over the DeepLabv3+ with the same backbone.

show abstract

“…After each convolution layer, a ReLu activation function exists to ensure non-linearity, which avoids acting like a single layer for sequential layers. The last layer in each block involves a 2×2 max-pooling that decreases the spatial resolution of feature maps (Das, Fime, Siddique, & Hashem, 2021). The filters of each convolution in the blocks are 64, 128, 256, 512, and 512.…”

Section: Vgg-16 As Backbonementioning

confidence: 99%

“…VGG-19 architecture contains five blocks with 16 convolution layers and has a small receptive field of 3x3 (Das et al, 2021). Each convolution layer follows a ReLu activation function, while a max-pooling is used in the last convolution layer of each block.…”

Section: Vgg-19 As Backbonementioning

confidence: 99%

Artificial Intelligence Based Instance-Aware Semantic Lobe Segmentation on Chest Computed Tomography Images

SAYRACI

AĞRALI

Kılıç

2022

EJOSAT

View full text Add to dashboard Cite

The coronavirus disease (COVID-19) has taken the entire world under its influence, causing a worldwide health crisis. The most concerning complication is acute hypoxemic respiratory failure that results in fatal consequences. To alleviate the effect of COVID-19, the infected region should be analyzed before the treatment. Thus, chest computed tomography (CT) is a popular method to determine the severity level of COVID-19. Besides, the number of lobe regions containing COVID-19 on CT images helps radiologists to diagnose the findings, such as bilateral, multifocal, and multilobar. Lobe regions can be distinguished manually by radiologists, but this may result in misdiagnosis due to human intervention. Therefore, in this study, a new tool has been developed that can automatically extract lobe regions using artificial intelligence-based instance-aware semantic lobe segmentation. Convolution neural networks (CNNs) offer automatic feature extraction in the instance-aware semantic lobe segmentation task that extracts the lobe regions on CT images. In this paper, CNN-based architectures, including DeepLabV3+ with VGG-16, VGG-19, and ResNet-50, were utilized to create a benchmark for the instance-aware semantic lobe segmentation task. For further improvement in segmentation results, images were preprocessed to detect the lung region prior to lobe segmentation. In the experimental evaluations, a large-scale dataset including 9036 images with pixel-level annotations for lung and lobe regions, has been created. DeepLabV3+ with ResNet-50 showed the highest performance in terms of dice similarity coefficient (DSC) and intersection over union (IOU) for lobe segmentation at 99.59 % and 99.19 %, respectively. The experiments demonstrated that our approach outperformed several state-of-the-art methods for the instance-aware semantic lobe segmentation task. Furthermore, a new desktop application called LobeChestApp was developed for the segmentation of lobe regions on chest CT images.

show abstract

Estimation of Road Boundary for Intelligent Vehicles Based on DeepLabV3+ Architecture

Cited by 29 publications

References 40 publications

mid-DeepLabv3+: A Novel Approach for Image Semantic Segmentation Applied to African Food Dietary Assessments

mid-DeepLabv3+: A Novel Approach for Image Semantic Segmentation Applied to African Food Dietary Assessments

mid-DeepLabv3+ : A Novel Approach for Image Semantic Segmentation applied to African Food Dietary Assessment

Artificial Intelligence Based Instance-Aware Semantic Lobe Segmentation on Chest Computed Tomography Images

Contact Info

Product

Resources

About