With the advancement of facial recognition technology, concerns over facial privacy breaches owing to data leaks and external attacks have been escalating. Existing de-identification methods face challenges with compa...
详细信息
With the advancement of facial recognition technology, concerns over facial privacy breaches owing to data leaks and external attacks have been escalating. Existing de-identification methods face challenges with compatibility with facial recognition models and difficulties in verifying de-identified images. To address these issues, this study introduces a novel framework that combines face verification-enabled de-identification techniques with face-swapping methods, tailored for video surveillance environments. This framework employs StyleGAN, Pixel2Style2Pixel (PSP), HopSkipJumpAttack (HSJA), and FaceNet512 to achieve face verification-capable de-identification, and uses the dlib library for face swapping. Experimental results demonstrate that this method maintains high face recognition performance (98.37%) across various facial recognition models while achieving effective de-identification. Additionally, human tests have validated its sufficient de-identification capabilities, and image quality assessments have shown its excellence across various metrics. Moreover, real-time de-identification feasibility was evaluated using Nvidia Jetson AGX Xavier, achieving a processing speed of up to 9.68 fps. These results mark a significant advancement in demonstrating the practicality of high-quality de-identification techniques and facial privacy protection in the field of video surveillance.
videomicroscopy is a promising tool combined with machine learning for studying the early development of in vitro fertilized bovine embryos and assessing its transferability as soon as possible. We aim to predict the ...
详细信息
ISBN:
(纸本)9798350349405;9798350349399
videomicroscopy is a promising tool combined with machine learning for studying the early development of in vitro fertilized bovine embryos and assessing its transferability as soon as possible. We aim to predict the embryo transferability within four days at most, taking 2D time-lapse microscopy videos as input. We formulate this problem as a supervised binary classification problem for the classes transferable and not transferable. The challenges are three-fold: 1) poorly discriminating appearance and motion, 2) class ambiguity, 3) small amount of annotated data. We propose a 3D convolutional neural network involving three pathways, which makes it multi-scale in time and able to handle appearance and motion in different ways. For training, we retain the focal loss. Our model, named SFR, compares favorably to other methods. Experiments demonstrate its effectiveness and accuracy for our challenging biological task.
Neural video codecs have recently become competitive with standard codecs such as HEVC in the low-delay setting. However, most neural codecs are large floating-point networks that use pixel-dense warping operations fo...
详细信息
ISBN:
(纸本)9798350318920;9798350318937
Neural video codecs have recently become competitive with standard codecs such as HEVC in the low-delay setting. However, most neural codecs are large floating-point networks that use pixel-dense warping operations for temporal modeling, making them too computationally expensive for deployment on mobile devices. Recent work has demonstrated that running a neural decoder in realtime on mobile is feasible, but shows this only for 720p RGB video. This work presents the first neural video codec that decodes 1080p YUV420 video in realtime on a mobile device. Our codec relies on two major contributions. First, we design an efficient codec that uses a block-based motion compensation algorithm available on the warping core of the mobile accelerator, and we show how to quantize this model to integer precision. Second, we implement a fast decoder pipeline that concurrently runs neural network components on the neural signal processor, parallel entropy coding on the mobile GPU, and warping on the warping core. Our codec outperforms the previous on-device codec by a large margin with up to 48 % BD-rate savings, while reducing the MAC count on the receiver side by 10x. We perform a careful ablation to demonstrate the effect of the introduced motion compensation scheme, and ablate the effect of model quantization.
There is a growing demand for real-timeimage denoising in low-light shooting with ultra-high definition cameras. This paper presents a denoising method that incorporates Haar-wavelet shrinkage denoising and a minimum...
详细信息
作者：
Lee, Seo-ElYoo, HyunChung, KyungyongKyonggi Univ
Dept Publ Safety Bigdata 154-42 Gwanggyosan Ro Suwon 16227 Gyeonggi Do South Korea Kyonggi Univ
Contents Convergence Software Res Ctr 154-42 Gwanggyosan Ro Suwon 16227 Gyeonggi Do South Korea Kyonggi Univ
Div AI Comp Sci & Engn 154-42 Gwanggyosan Ro Suwon 16227 Gyeonggi Do South Korea
Capitalizing on the rapid development of diverse deep learning technologies in the field of image analysis, studies are now being conducted to detect objects within images, estimate the poses of target objects, and cl...
详细信息
Capitalizing on the rapid development of diverse deep learning technologies in the field of image analysis, studies are now being conducted to detect objects within images, estimate the poses of target objects, and classify motions. However, the magnitude of image data and computational complexity present challenges in performing real-timeimage analysis. In addition, the classification of human motions, specifically, requires an effective methodology based on analysis of the changing features of poses from frame to frame. To address this, pose pattern mining using a transformer for motion classification is proposed, which expands the output of the neural network of an object detection model into a time-series pose classification model combining EfficientNet and transformer mechanisms. With regard to the structure of the configured model, the object detection model maintains the mechanisms and displays the effect of expanding the output node of the internal neural network into a time-series-based pose-motion analysis model. Sequence pattern mining is then applied to ensure the efficiency of the data analysis. This hybrid methodology achieved a response time of 0.059 s per frame and an accuracy of 86.67%. Therefore, it may be surmised that this proposed method can be applied to fields such as security and surveillance systems that require fast processingtimes and high levels of accuracy.
real-timeimageprocessing methods are well applied in engineering practice. This study presents a breakthrough real-timeimage transmission algorithm designed specifically for remote electronic engineering laboratori...
详细信息
video synthetic aperture radar (video SAR) has become a research hotspot in the SAR field due to its characteristic of continuous monitoring. Compared to traditional SAR, video SAR provides the ability to dynamically ...
详细信息
With the rapid development of artificial intelligence technology, deep learning has become one of the key technologies in the field of image recognition. PyTorch has become the preferred framework for researchers due ...
详细信息
video target tracking has a wide range of application value in the field of automatic driving, UAV target tracking, security monitoring, etc. How to maintain stable tracking of the target among video data frames is th...
详细信息
Interactive information fault diagnosis technology is a new type of fault diagnosis technology which is integrated by information fusion, artificial intelligence, computer science and other disciplines. It can extract...
详细信息
暂无评论