A Comprehensive Review of YOLO Architectures in Computer Vision: From YOLOv1 to YOLOv8 and YOLO-NAS

Terven, Juan; Córdova-Esparza, Diana-Margarita; Romero-González, Julio-Alejandro

doi:10.3390/make5040083

Cited by 649 publications

(170 citation statements)

References 113 publications

Supporting

Mentioning

169

Contrasting

Unclassified

Order By: Relevance

“…2a). 12,13 The detection model output a bounding box for the bAVM. In the second stage, an image masked by the bounding box was input into the U-Net and U-Net++ architecture, and a segmentation mask was outputted (Fig.…”

Section: Data Preprocessing and Deep Learning Approachesmentioning

confidence: 99%

“…YOLOv5x and YOLOv8x, which had the best performance but a relatively long inference time, were used. 12,13 The models with pretraining and image augmentation methods were compared to the model without those methods. Additionally, two types of pretrained weights were tested: the weight training by the COCO (Common Objects In Context) dataset provided by YOLOv5 and YOLOv8 originally and the weight training by the MRI images with YOLOv5 to detect brain tumors (Fig.…”

Section: Yolov5 and Yolov8 Configurationmentioning

confidence: 99%

“…The inferred bounding box and the ground-truth bounding box should have an IoU ≥0.5 to be considered the correct inference for an IoU of 0.5. 12,13 Those metrics can be found in GitHub for this study and for YOLOv5 and YOLOv8. 12,13 The Dice coefficient and balanced average Hausdorff distance in percentage were used to evaluate the performance for the contour of the segmentation.…”

Section: Performance Metricsmentioning

confidence: 99%

“…12,13 Those metrics can be found in GitHub for this study and for YOLOv5 and YOLOv8. 12,13 The Dice coefficient and balanced average Hausdorff distance in percentage were used to evaluate the performance for the contour of the segmentation. 18,19 In the balanced average Hausdorff distance, it was factored in segmentation size, which was absent in the original formula.…”

Section: Performance Metricsmentioning

confidence: 99%

“…11 YOLOv5 and YOLOv8 are developed by the same group, and both are state-of-theart object detection algorithms in computer vision. 12,13 Both YOLOv5 and YOLOv8 have been widely adopted in various applications, such as self-driving cars, video surveillance, and object tracking. For medical image segmentation, U-Net++ is an improved version of the U-Net architecture that can fully use multiscale feature maps and outperform U-Net in many tasks.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Deep Learning Detection and Segmentation of Brain Arteriovenous Malformation on Magnetic Resonance Angiography

Hong

You

Sun

et al. 2023

Magnetic Resonance Imaging

View full text Add to dashboard Cite

BackgroundThe delineation of brain arteriovenous malformations (bAVMs) is crucial for subsequent treatment planning. Manual segmentation is time‐consuming and labor‐intensive. Applying deep learning to automatically detect and segment bAVM might help to improve clinical practice efficiency.PurposeTo develop an approach for detecting bAVM and segmenting its nidus on Time‐of‐flight magnetic resonance angiography using deep learning methods.Study TypeRetrospective.Subjects221 bAVM patients aged 7–79 underwent radiosurgery from 2003 to 2020. They were split into 177 training, 22 validation, and 22 test data.Field Strength/Sequence1.5 T, Time‐of‐flight magnetic resonance angiography based on 3D gradient echo.AssessmentThe YOLOv5 and YOLOv8 algorithms were utilized to detect bAVM lesions and the U‐Net and U‐Net++ models to segment the nidus from the bounding boxes. The mean average precision, F1, precision, and recall were used to assess the model performance on the bAVM detection. To evaluate the model's performance on nidus segmentation, the Dice coefficient and balanced average Hausdorff distance (rbAHD) were employed.Statistical TestsThe Student's t‐test was used to test the cross‐validation results (P < 0.05). The Wilcoxon rank test was applied to compare the median for the reference values and the model inference results (P < 0.05).ResultsThe detection results demonstrated that the model with pretraining and augmentation performed optimally. The U‐Net++ with random dilation mechanism resulted in higher Dice and lower rbAHD, compared to that without that mechanism, across varying dilated bounding box conditions (P < 0.05). When combining detection and segmentation, the Dice and rbAHD were statistically different from the references calculated using the detected bounding boxes (P < 0.05). For the detected lesions in the test dataset, it showed the highest Dice of 0.82 and the lowest rbAHD of 5.3%.Data ConclusionThis study showed that pretraining and data augmentation improved YOLO detection performance. Properly limiting lesion ranges allows for adequate bAVM segmentation.Level of Evidence4Technical Efficacy Stage1

show abstract

Section: Data Preprocessing and Deep Learning Approachesmentioning

confidence: 99%

Section: Yolov5 and Yolov8 Configurationmentioning

confidence: 99%

Section: Performance Metricsmentioning

confidence: 99%

Section: Performance Metricsmentioning

confidence: 99%

mentioning

confidence: 99%

See 3 more Smart Citations

Deep Learning Detection and Segmentation of Brain Arteriovenous Malformation on Magnetic Resonance Angiography

Hong

You

Sun

et al. 2023

Magnetic Resonance Imaging

View full text Add to dashboard Cite

show abstract

A Silver Nanowires‐Based Flexible Capacitive Touch Screen in Tactile Displays for Individuals with Visual Impairment Using Gesture Recognition

Hamza,

Alzalabny,

Buduru

et al. 2024

Adv Materials Technologies

View full text Add to dashboard Cite

Capacitive touch screens (CTS's) are essential components in most of today's digital devices. However, for the visually impaired (VI) users due to the uneven topography of the tactile surface, CTS's are more challenging to implement and thus this field remains largely underdeveloped. Considering the limited space around the microactuators driving the typical Braille dots for a tactile screen with ten dots‐per‐inch (dpi) resolution, the materials used for CTS should be flexible and durable with high mechanical strength. In this work, a flexible CTS based on polyimide (PI) and silver nanowires (AgNWs) as electrodes with a total thickness of 210 µm is developed. The dimensions of the AgNWs are on average 7.9 ± 2.4 µm in length and 85 ± 24 nm in width. The AgNWs electrodes showed low resistance and good adhesion to the PI substrate. A gesture recognition application is collected from the capacitive data to classify different gestures (including single‐ and double‐click, swipe‐left and ‐right, scroll‐up and ‐down as well as zoom‐in and ‐out) with two different approaches; machine learning and deep learning are implemented. The best performance is obtained using the YOLO model with a high validation accuracy of 97.76%. Finally, a software application is developed with the proposed hand gestures in real‐time to foster interaction of VI users with the tactile display allowing them to navigate a Windows file system and interact with the documents via hand gestures in a similar manner as sighted users on a conventional touch display will be able to do. This work paves the way to utilize CTS for the tactile displays in the market developed for VI users.

show abstract

Research on Fine‐Grained Phenotypic Analysis of Temporal Root Systems – Improved YoloV8seg Applied for Fine‐Grained Analysis of In Situ Root Temporal Phenotypes

Yu,

Zhang,

Wang

et al. 2024

Advanced Science

View full text Add to dashboard Cite

Root systems are crucial organs for crops to absorb water and nutrients. Conducting phenotypic analysis on roots is of great importance. To date, methods for root system phenotypic analysis have predominantly focused on semantic segmentation, integrating phenotypic extraction software to achieve comprehensive root phenotype analysis. This study demonstrates the feasibility of instance segmentation tasks on in situ root system images. An improved YoloV8n‐seg network tailored for detecting elongated roots is proposed, which outperforms the original YoloV8seg in all network performance metrics. Additionally, the post‐processing method introduced reduces root identification errors, ensuring a one‐to‐one correspondence between each root system and its detection box. The experiment yields phenotypic parameters for fine‐grained roots, such as fine‐grained root length, diameter, and curvature. Compared to traditional parameters like total root length and average root diameter, these detailed phenotypic analyses enable more precise phenotyping and facilitate accurate artificial intervention during crop cultivation.

show abstract

A Comprehensive Review of YOLO Architectures in Computer Vision: From YOLOv1 to YOLOv8 and YOLO-NAS

Cited by 649 publications

References 113 publications

Deep Learning Detection and Segmentation of Brain Arteriovenous Malformation on Magnetic Resonance Angiography

Deep Learning Detection and Segmentation of Brain Arteriovenous Malformation on Magnetic Resonance Angiography

A Silver Nanowires‐Based Flexible Capacitive Touch Screen in Tactile Displays for Individuals with Visual Impairment Using Gesture Recognition

Research on Fine‐Grained Phenotypic Analysis of Temporal Root Systems – Improved YoloV8seg Applied for Fine‐Grained Analysis of In Situ Root Temporal Phenotypes

Contact Info

Product

Resources

About