Neural Attention-Driven Non-Maximum Suppression for Person Detection

Symeonidis, Charalampos; Mademlis, Ioannis; Pitas, Ioannis; Nikolaidis, Nikos

doi:10.1109/tip.2023.3268561

Cited by 11 publications

(3 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Soft-NMS is an improved loss function used to suppress redundant bounding boxes more smoothly during the NMS process [30]. Previously, the traditional NMS method used a fixed threshold to determine if two bounding boxes overlap and to suppress them [31]. However, due to the fixed threshold, it may adapt to situations between different objects, resulting in some candidate boxes with low confidence but which incorrectly exclude significantly overlapping true objects.…”

Section: Soft-nms Loss Functionmentioning

confidence: 99%

SGST-YOLOv8: An Improved Lightweight YOLOv8 for Real-Time Target Detection for Campus Surveillance

Cheng,

Chao,

Yang

et al. 2024

Applied Sciences

View full text Add to dashboard Cite

Real-time target detection plays an important role in campus intelligent surveillance systems. This paper introduces Soft-NMS, GSConv, Triplet Attention, and other advanced technologies to propose a lightweight pedestrian and vehicle detection model named SGST-YOLOv8. In this paper, the improved YOLOv8 model is trained on the self-made dataset, and the tracking algorithm is combined to achieve an accurate and efficient real-time pedestrian and vehicle tracking detection system. The improved model achieved an accuracy of 88.6%, which is 1.2% higher than the baseline model YOLOv8. Additionally, the mAP0.5:0.95 increased by 3.2%. The model parameters and GFLOPS reduced by 5.6% and 7.9%, respectively. In addition, this study also employed the improved YOLOv8 model combined with the bot sort tracking algorithm on the website for actual detection. The results showed that the improved model achieves higher FPS than the baseline YOLOv8 model when detecting the same scenes, with an average increase of 3–5 frames per second. The above results verify the effectiveness of the improved model for real-time target detection in complex environments.

show abstract

Section: Soft-nms Loss Functionmentioning

confidence: 99%

SGST-YOLOv8: An Improved Lightweight YOLOv8 for Real-Time Target Detection for Campus Surveillance

Cheng,

Chao,

Yang

et al. 2024

Applied Sciences

View full text Add to dashboard Cite

show abstract

“…When predicting the signal positions for candidate regions, the model generates multiple prediction boxes, which inevitably raises the question of how to eliminate redundant predictions. Most approaches rely on post-processing steps, such as non-maximum suppression (NMS) [8] or its related improved algorithms [9], to filter and retain the most probable box among multiple potential predictions for the same target by setting a threshold. However, there is significant variation in the aspect ratios of signals in the spectrogram, making it challenging for well-designed prior anchors to match different signals.…”

Section: Introductionmentioning

confidence: 99%

Intelligent Detection Method for Satellite TT&C Signals under Restricted Conditions Based on TATR

Li,

Shi,

Wang

et al. 2024

Remote Sensing

View full text Add to dashboard Cite

In complex electromagnetic environments, satellite telemetry, tracking, and command (TT&C) signals often become submerged in background noise. Traditional TT&C signal detection algorithms suffer a significant performance degradation or can even be difficult to execute when phase information is absent. Currently, deep-learning-based detection algorithms often rely on expert-experience-driven post-processing steps, failing to achieve end-to-end signal detection. To address the aforementioned limitations of existing algorithms, we propose an intelligent satellite TT&C signal detection method based on triplet attention and Transformer (TATR). TATR introduces the residual triplet attention (ResTA) backbone network, which effectively combines spectral feature channels, frequency, and amplitude dimensions almost without introducing additional parameters. In signal detection, TATR employs a multi-head self-attention mechanism to effectively address the long-range dependency issue in spectral information. Moreover, the prediction-box-matching module based on the Hungarian algorithm eliminates the need for non-maximum suppression (NMS) post-processing steps, transforming the signal detection problem into a set prediction problem and enabling parallel output of the detection results. TATR combines the global attention capability of ResTA with the local self-attention capability of Transformer. Experimental results demonstrate that utilizing only the signal spectrum amplitude information, TATR achieves accurate detection of weak TT&C signals with signal-to-noise ratios (SNRs) of −15 dB and above (mAP@0.5 > 90%), with parameter estimation errors below 3%, which outperforms typical target detection methods.

show abstract

“…The input plane receives images of characters with a standardized size and centeredness, and each layer's units receive input from a set of smaller units located in the previous layer [11]. Two unique attributes of the architecture of CNNs are their sparse connections and shared weights [12]. CNNs exploit spatially local correlation by imposing a pattern of local connectivity between neurons of adjacent layers, where units in layer m are connected to three spatially adjacent units in layer m − 1.…”

Section: Introductionmentioning

confidence: 99%

Model Development for Identifying Aromatic Herbs Using Object Detection Algorithm

Antunes,

Okano,

Nääs

et al. 2024

AgriEngineering

View full text Add to dashboard Cite

The rapid evolution of digital technology and the increasing integration of artificial intelligence in agriculture have paved the way for groundbreaking solutions in plant identification. This research pioneers the development and training of a deep learning model to identify three aromatic plants—rosemary, mint, and bay leaf—using advanced computer-aided detection within the You Only Look Once (YOLO) framework. Employing the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology, the study meticulously covers data understanding, preparation, modeling, evaluation, and deployment phases. The dataset, consisting of images from diverse devices and annotated with bounding boxes, was instrumental in the training process. The model’s performance was evaluated using the mean average precision at a 50% intersection over union (mAP50), a metric that combines precision and recall. The results demonstrated that the model achieved a precision of 0.7 or higher for each herb, though recall values indicated potential over-detection, suggesting the need for database expansion and methodological enhancements. This research underscores the innovative potential of deep learning in aromatic plant identification and addresses both the challenges and advantages of this technique. The findings significantly advance the integration of artificial intelligence in agriculture, promoting greater efficiency and accuracy in plant identification.

show abstract

Neural Attention-Driven Non-Maximum Suppression for Person Detection

Cited by 11 publications

References 44 publications

SGST-YOLOv8: An Improved Lightweight YOLOv8 for Real-Time Target Detection for Campus Surveillance

SGST-YOLOv8: An Improved Lightweight YOLOv8 for Real-Time Target Detection for Campus Surveillance

Intelligent Detection Method for Satellite TT&C Signals under Restricted Conditions Based on TATR

Model Development for Identifying Aromatic Herbs Using Object Detection Algorithm

Contact Info

Product

Resources

About