Internet Images Research Articles

Federated Learning on Heterogeneous Images in Internet of Biomedical Things

Federated Learning on Heterogeneous Images in Internet of Biomedical Things

Metric3D v2: A Versatile Monocular Geometric Foundation Model for Zero-Shot Metric Depth and Surface Normal Estimation.

We introduce Metric3D v2, a geometric foundation model designed for zero-shot metric depth and surface normal estimation from single images, critical for accurate 3D recovery. Depth and normal estimation, though complementary, present distinct challenges. State-of-the-art monocular depth methods achieve zero-shot generalization through affine-invariant depths, but fail to recover real-world metric scale. Conversely, current normal estimation techniques struggle with zero-shot performance due to insufficient labeled data. We propose targeted solutions for both metric depth and normal estimation. For metric depth, we present a canonical camera space transformation module that resolves metric ambiguity across various camera models and large-scale datasets, which can be easily integrated into existing monocular models. For surface normal estimation, we introduce a joint depth-normal optimization module that leverages diverse data from metric depth, allowing normal estimators to improve beyond traditional labels. Our model, trained on over 16 million images from thousands of camera models with varied annotations, excels in zero-shot generalization to new camera settings. As shown in Fig. 1, It ranks the 1st in multiple zero-shot and standard benchmarks for metric depth and surface normal prediction. Our method enables the accurate recovery of metric 3D structures on randomly collected internet images, paving the way for plausible single-image metrology. Our model also relieves the scale drift issues of monocular-SLAM (Fig. 3), leading to high-quality metric scale dense mapping. Such applications highlight the versatility of Metric3D v2 models as geometric foundation models.

Publisher Correction: Internet image search outputs propagate climate change sentiment and impact policy support

Publisher Correction: Internet image search outputs propagate climate change sentiment and impact policy support

Internet image search outputs propagate climate change sentiment and impact policy support

AbstractA critical step in tackling climate change involves structural, system-level changes facilitating action. Despite their ubiquity, little is known about how internet search algorithms portray climate change, and how these portrayals impact concern and action. In a sample of 49 countries, we found that nationwide climate concern, but not nation-level climate impact, predicted the emotional arousal caused by climate change Google Image Search outputs, as rated by a naive sample (n = 383). In a follow-up experiment we randomly assigned another sample (n = 899) to receive the climate change image outputs resulting from searches conducted in countries high or low in pre-existing climate concern, and found that participants exposed to images from countries with high pre-existing concern (compared to low) became more concerned about climate change, supportive of climate policy and likely to act pro-environmentally, suggesting a cycle of climate sentiment propagation systemically facilitated by internet search algorithms. We discuss the implications of these findings for climate action interventions.

Identification of pedestrian submerged parts in urban flooding based on images and deep learning

During urban flooding, pedestrians are often trapped in floodwater, and some pedestrians even fall or drown. The pedestrian submerged part (i.e., the human body part that water surface reaches) is an important reference indicator for judging dangerous situation of pedestrians. Flood images usually contain the information about pedestrian submerged parts. We proposed an automated method for identifying pedestrian submerged parts from images. This method utilizes relevant deep learning technologies to segment water surfaces, detect the pedestrians in floodwater, and detect the human keypoints of the pedestrians from images, and then identify submerged parts of the pedestrians according to the relationship between the human keypoints and the water surfaces. This method achieves an accuracy of 90.71% in identifying pedestrian submerged parts on an image dataset constructed from Internet images. The result shows that this method could effectively identify pedestrian submerged parts from images with high accuracy.

Deep Learning Method for Real-Time Fire Detection System for Urban Fire Monitoring and Control

During urban fire incidents, real-time videos and images are vital for emergency responders and decision-makers, facilitating efficient decision-making and resource allocation in smart city fire monitoring systems. However, real-time videos and images require simple and embeddable models in small computer systems with highly accurate fire detection ratios. YOLOv5s has a relatively small model size and fast processing time with limited accuracy. The aim of this study is to propose a method that employs a YOLOv5s network with a squeeze-and-excitation module for image filtering and classification to meet the urgent need for rapid and accurate real-time screening of irrelevant data. In this study, over 3000 internet images were used for crawling and annotating to construct a dataset. Furthermore, the YOLOv5, YOLOv5x and YOLOv5s models were developed to train and test the dataset. Comparative analysis revealed that the proposed YOLOv5s model achieved 98.2% accuracy, 92.5% recall, and 95.4% average accuracy, with a remarkable processing speed of 0.009 s per image and 0.19 s for a 35 frames-per-second video. This surpasses the performance of other models, demonstrating the efficacy of the proposed YOLOv5s for real-time screening and classification in smart city fire monitoring systems.

The Metaverse in Green Building Concept Learning, Creative Design Performance, and Learning Engagement

Due to the rapid advancement of technology, environmental sustainability education has gained significant attention. This study aimed to explore the effect of the metaverse on green building concept learning, creative design performance, and learning engagement. This research was conducted with 61 students from a public high school, employing a quasi-experimental design with nonequivalent groups. The experimental teaching theme was green building education, where the experimental group used online metaverse with Minecraft-assisted teaching, while the comparison group used internet images for assistance. Our research results were as follows: 1. The metaverse had a large effect size on the example dimension of concept learning but no significant effect on the linkages, hierarchical structures, or cross-linking terms. 2. The metaverse significantly enhanced creative design performance, including value, usability, novelty, and elaboration. 3. The metaverse had a significant effect on learning engagement, particularly in learning attitudes.

Effectiveness of digital marketing and its value in new ventures

This paper analyzes the effect of digital marketing and cluster membership on the entrepreneurship and performance of dental clinics in Colombia. The use of different digital marketing tools in dental clinics is evaluated, and the effect of these tools on the creation of new markets, and clinics’ services and general performance, are analyzed. In addition, the cluster effect on the learning and implementation of digital tools in individual clinics is studied. The results show the most effective digital tools to create new markets, as well as the tools that improve clinics’ performance through the Internet image and online visibility. The combined use of corporate Facebook and Instagram, a robust website with wide online visibility and YouTube develops the image of companies on the Internet in the most effective way. In addition, the importance of belonging to a cluster in the learning and implementation of digital tools in individual clinics was verified.

Tools Identification By On-Board Adaptation of Vision-and-Language Models

A robotic workshop assistant has been a long-standing grand challenge for robotics, speech, computer vision, and artificial intelligence (AI) research. We revisit the goal of visual identification of tools from human queries in the current era of Large Vision-and-Language models (like GPT-4). We find that current off-the-shelf models (that are trained on internet images) are unable to overcome the domain shift and unable to identify small, obscure tools in cluttered environments. Furthermore, these models are unable to match tools to their intended purpose or affordances. We present a novel system for online domain adaptation that can be run directly on a small on-board processor. The system uses Hyperdimensional Computing (HD), a fast and efficient neuromorphic method. We adapted CLIP to work with explicit ("I need the hammer") and implicit purpose-driven queries ("Drive these nails"), and even with depth images as input. This demo allows the user to try out various real tools and interact via free-form audio.

Surviving covid-19: a multimodal discourse analysis of new media covid-19 vaccination-related pictorials

Abstract This paper explores the communicative acts deployed in covid-19 vaccination-related pictorials circulated on digital media platforms. Seven internet images were purposively sampled with a view to exploring their communicative functions as well as their generic structure. The data, which were culled from the websites of the World Health Organisation, Centre for Disease Control, Pan American Health Organisation and Facebook, were subjected to qualitative analysis. The study deployed van Leeuwen’s Multimodal Discourse Analysis and Yuen’s Generic Structure Potential as theoretical anchor. The multimodal communicative acts are deployed for instructive, illustrative, informative, persuasive, inviting and advisory purposes. Categories such as Lead, Emblem, Announcement and Enhancer are compulsory in the data while Display, Tag and Call-and-Visit Information are non-compulsory elements. This can be catalogued as: ‘Lead^(Display)^Emblem^(Announcement)^(Enhancer)^(Tag)^(Call-and-Visit Information)’. The study contends that the various semiotic resources deployed in the internet-circulated covid-19 images are used not only for informative and other communicative purposes but also to evoke attitudinal change towards and encourage widespread acceptance of the covid-19 vaccines.

Automatic Food Recognition Using Deep Convolutional Neural Networks with Self-attention Mechanism

The significance of food in human health and well-being cannot be overemphasized. Nowadays, in our dynamic life, people are increasingly concerned about their health due to increased nutritional ailments. For this reason, mobile food-tracking applications that require a reliable and robust food classification system are gaining popularity. To address this, we propose a robust food recognition model using deep convolutional neural networks with a self-attention mechanism (FRCNNSAM). By training multiple FRCNNSAM structures with varying parameters, we combine their predictions through averaging. To prevent over-fitting and under-fitting data augmentation to generate extra training data, regularization to avoid excessive model complexity was used. The FRCNNSAM model is tested on two novel datasets: Food-101 and MA Food-121. The model achieved an impressive accuracy of 96.40% on the Food-101 dataset and 95.11% on MA Food-121. Compared to baseline transfer learning models, the FRCNNSAM model surpasses performance by 8.12%. Furthermore, the evaluation on random internet images demonstrates the model's strong generalization ability, rendering it suitable for food image recognition and classification tasks.

Combining Riemann–Lebesgue-Based Key Generator and Machine-Learning-Based Intelligent Encryption Scheme for IoMT Images Infosecurity

Imaging examinations, such as chest X-ray (CXR) images, can be used for early detection of diseases inside the human body, such as lung cancer, and cardiopulmonary-related diseases; however, they may contain patients’ private information, which raises concerns about the security levels of medical images in internet of medical things (IoMT) systems. The promising authorization protocol algorithm in 5th-generation networking technology can protect medical image confidentiality, recoverability, and availability. Hence, we propose a symmetric cryptography protocol based on the Riemann-Lebesgue (R-L) -based key generator and a machine learning (ML)-based cryptography scheme for CXR image infosecurity. The R-L-based key generator uses the composite function“, g(x)sin((1/2+N)x)”, to dynamically produce the oscillation function by controlling the terms of the nonlinear function“, g(x)”, and frequency“, N”, which are used to produce the randomnumber seed and to select non-ordered numbers and nonrepeated 256 secret keys in the key space. The ML-based scheme uses these random secret keys to train an image encryptor and decryptor, respectively. Based on the diffusion method, the ML-based cryptography scheme is used to change image pixel values. For evaluation experiments, CXR images are collected from the National Institutes of Health Chest X-ray database and used to evaluate image quality after encryption and decryption processes. For different cardiopulmonary-related disease images, the number of pixel change rates (NPCR) and unified averaged changed intensity (UACI) are used to evaluate the security level and decryption quality. The experimental results show an average NPCR of 99.6% and average UACI of 32.61% between plain and cipher images for evaluating the security level, enabling further online diagnosis applications in IoMT systems.

EC-SfM: Efficient Covisibility-Based Structure-From-Motion for Both Sequential and Unordered Images

Structure-from-Motion is a technology used to obtain scene structure through image collection, which is a fundamental problem in computer vision. For unordered Internet images, SfM is very slow due to the lack of prior knowledge about image overlap. For sequential images, knowing the large overlap between adjacent frames, SfM can adopt a variety of acceleration strategies, which are only applicable to sequential data. To further improve the reconstruction efficiency and break the gap of strategies between these two kinds of data, this paper presents an efficient covisibility-based incremental SfM. Different from previous methods, we exploit covisibility and registration dependency to describe the image connection which is suitable to any kind of data. Based on this general image connection, we propose a unified framework to efficiently reconstruct sequential images, unordered images, and the mixture of these two. Experiments on the unordered images and mixed data verify the effectiveness of the proposed method, which is three times faster than the state-of-the-art on feature matching, and an order of magnitude faster on reconstruction without sacrificing the accuracy. The source code is publicly available at https://github.com/openxrlab/xrsfm.

Development of Android-Based 3D Building Augmented Reality Application at Jogja Student Islamic Boarding School 2

Islamic boarding schools are pivotal in providing religious education for Muslims in Indonesia. Despite their widespread presence, detailed knowledge about these institutions, particularly their architectural design, remains limited. Prospective students and the public often rely on internet images for information, leading to a partial understanding of the physical structure of these schools. This study introduces an innovative solution through the development of an Augmented Reality (AR) mobile application specifically designed for the Jogja 2 Islamic Boarding School. Augmented Reality technology effectively integrates 2D and 3D virtual objects into a real-world environment, offering an immersive, real-time experience. By utilizing the waterfall methodology in its development, this application aims to provide an in-depth, three-dimensional exploration of the Jogja 2 boarding school's infrastructure. This tool is intended to assist potential students and interested individuals in gaining a comprehensive understanding of the school's architectural layout and design dynamically and interactively.

Improved deep semantic dictionary learning for multi-label image classification

Multi-label image classification is a practical and challenging task in the field of machine learning. It is a fundamental but essential task used in auto driving, internet image classification, and other fields. Many researchers have done various researches on relevant aspects. Deep Semantic Dictionary Learning (DSDL) is a multi-label image classification method brought out by previous researchers which combines the semantic meaning information into the classification task. However, it only uses one kind of feature extraction network as the feature map generator without considering other alternatives. In this paper, six types of pretrained networks are applied as alternative choices to validate the performance of DSDL on different neural networks. Experiments are conducted on the VOC2007 dataset and the result shows the inner relation of different series of neural networks and demonstrates that different feature extraction networks perform variously on different tasks. The ResNet-152 achieves the mean average precision (mAP) of 94.5% and outperforms the original DSDL method of 0.3% mAP.

Color Polymorphism of Unfeathered Parts of the Great White Egret Ardea alba alba in Europe: An Analysis of the Internet Images

Polymorphism has been described in many avian taxa including Ardeidae. This study investigated the spatio-temporal distribution of color polymorphic forms (with regard to the color of the unfeathered parts (legs/lores)) of the Great White Egret Ardea alba alba in internet images from Europe. In total, 140 images of presumed polymorphs from 16 countries (from 2006 to 2021) have been collected and examined. Eight coloration schemes of visible unfeathered parts were distinguished. Since some polymorphs may represent the subspecies Eastern Great Egret Ardea alba modesta, the ratio of beak length to total head length was compared among individuals representing the nominal and Eastern Great Egret subspecies. The ratio distributions were similar in all compared groups; thus, polymorph egrets may represent both of these subspecies. Most polymorph images originate from the Netherlands and the southwestern Norwegian coast. This may have resulted from specific local diets and contact with hormone-disrupting compounds. Nevertheless, some observation biases, such as the presence of local breeding population or high numbers of nature photographers, could have affected our observation pattern. Some analyzed features also followed temporal patterns. Most polymorphs were observed during the breeding season, suggesting disruptions of sex hormones as a possible cause. Studies of spatio-temporal patterns of polymorph occurrence may allow us to obtain insights into the causes of large variation in coloration schemes in polymorphic species.

Generation of a Melanoma and Nevus Data Set From Unstandardized Clinical Photographs on the Internet

Artificial intelligence (AI) training for diagnosing dermatologic images requires large amounts of clean data. Dermatologic images have different compositions, and many are inaccessible due to privacy concerns, which hinder the development of AI. To build a training data set for discriminative and generative AI from unstandardized internet images of melanoma and nevus. In this diagnostic study, a total of 5619 (CAN5600 data set) and 2006 (CAN2000 data set; a manually revised subset of CAN5600) cropped lesion images of either melanoma or nevus were semiautomatically annotated from approximately 500 000 photographs on the internet using convolutional neural networks (CNNs), region-based CNNs, and large mask inpainting. For unsupervised pretraining, 132 673 possible lesions (LESION130k data set) were also created with diversity by collecting images from 18 482 websites in approximately 80 countries. A total of 5000 synthetic images (GAN5000 data set) were generated using the generative adversarial network (StyleGAN2-ADA; training, CAN2000 data set; pretraining, LESION130k data set). The area under the receiver operating characteristic curve (AUROC) for determining malignant neoplasms was analyzed. In each test, 1 of the 7 preexisting public data sets (total of 2312 images; including Edinburgh, an SNU subset, Asan test, Waterloo, 7-point criteria evaluation, PAD-UFES-20, and MED-NODE) was used as the test data set. Subsequently, a comparative study was conducted between the performance of the EfficientNet Lite0 CNN on the proposed data set and that trained on the remaining 6 preexisting data sets. The EfficientNet Lite0 CNN trained on the annotated or synthetic images achieved higher or equivalent mean (SD) AUROCs to the EfficientNet Lite0 trained using the pathologically confirmed public data sets, including CAN5600 (0.874 [0.042]; P = .02), CAN2000 (0.848 [0.027]; P = .08), and GAN5000 (0.838 [0.040]; P = .31 [Wilcoxon signed rank test]) and the preexisting data sets combined (0.809 [0.063]) by the benefits of increased size of the training data set. The synthetic data set in this diagnostic study was created using various AI technologies from internet images. A neural network trained on the created data set (CAN5600) performed better than the same network trained on preexisting data sets combined. Both the annotated (CAN5600 and LESION130k) and synthetic (GAN5000) data sets could be shared for AI training and consensus between physicians.

Accuracy and Utility of Internet Image Search as a Learning Tool for Retinal Pathology.

Purpose Ophthalmology residency training heavily relies on visual and pattern recognition-based learning. In parallel with traditional reference texts, online internet search via Google Image Search (GIS) is commonly used and offers an accessible fund of reference images for ophthalmology trainees seeking rapid exposure to images of retinal pathology. However, the accuracy and quality of this tool within this context is unknown. We aim to evaluate the accuracy and quality of GIS images of selected retinal pathologies. Methods A cross-sectional study was performed of GIS of 15 common and 15 rare retinal diseases drawn from the American Academy of Ophthalmology residency textbook series. A total of 300 evaluable image results were assessed for accuracy of images and image source accountability in consultation with a vitreoretinal surgeon. Results A total of 377 images were reviewed with 77 excluded prior to final analysis. A total of 288 (96%) search results accurately portrayed the retinal disease being searched, whereas 12 (4%) were of an erroneous diagnosis. More images of common retinal diseases were from patient education Web sites than were images of rare diseases ( p < 0.01). Significantly more images of rare retinal diseases were found in peer-reviewed sources ( p = 0.01). Conclusions GIS search results yielded a modest level of accuracy for the purposes of ophthalmic education. Despite the ease and rapidity of accessing multimodal retinal imaging examples, this tool may best be suited as a supplementary resource for learning among residents due to limited accuracy, lack of sufficient supporting information, and the source Web site's focus on patient education.

An Automated Learning Framework With Limited and Cross-Domain Data for Traffic Equipment Detection From Surveillance Videos

Traffic equipment detection from surveillance videos is of practical significance for temporary traffic element update in high-precision maps. However, there is little relative research developed due to limited labeled data. Based on a detector dubbed Faster R-CNN, we propose an automated learning framework that utilizes easy-to-obtain Internet images containing traffic equipment to acquire the capability of detecting traffic equipment from surveillance videos. In this framework, an appearance weighting module using a comprehensive feature aggregation method is designed to allow Faster R-CNN to converge and generalize quickly by taking limited data (i.e., less than 30 images per class) as input. To further address the cross-domain issue brought by the domain gap between the Internet images and the surveillance video frames, a domain adaptation learning scheme is developed, which aims to align the two domains and guide the framework to learn more robust domain-invariant features. Experimental results show that both the appearance weighting module and the domain adaptation learning scheme could bring a great performance improvement. Moreover, the combination of the two results in a state-of-the-art performance (mAP of 44.6%) even if only 30 training images per class are provided. To sum up, the proposed framework is suitable for traffic equipment detection from surveillance videos and provides an inspiration for other detection tasks with limited and cross-domain data, allowing humans to reduce their efforts and time required for arduous data collection and annotation.

Application Research of Virtual Reality Animation Technology in Agricultural Literature Creation Image Internet of Things

As an important part of Huxiang culture, Meishan’s regional culture has grown independently in a semiclosed environment for a long time, so it has become a relatively complete and ancient mysterious form of culture. However, due to the scattered distribution of tourism resources, there are many difficulties in the research and development of Meishan culture. Based on the characteristics of Zhang Wulang’s regional culture, this paper uses the theory of virtual reality technology to guide research and design practice. First, the definitions of virtual tourism, virtual reality technology, Meishan culture and Zhang Wulang woodcarving techniques, artistic modelling, and engraving style are defined, and several other technical features and traits are discussed. Secondly, the laser scanning technology was used to carry out three-dimensional modelling of the characteristic buildings in Meishan regional culture and the representative Zhang Wulang woodcarving works. Finally, the characteristics of Meishan cultural tourism resources are combined to comprehensively grasp the development of tourism resources, the three-dimensional image reconstruction of Zhang Wulang model, and the dissemination of multimedia forms.