Refine
Document Type
- Conference proceeding (26)
- Journal article (15)
Language
- English (41)
Is part of the Bibliography
- yes (41)
Institute
- Technik (40)
- Informatik (2)
- Texoversum (2)
- Zentrale Einrichtungen (1)
Publisher
- IEEE (9)
- Springer (7)
- SCITEPRESS (4)
- ACM (2)
- International Academy Publishing (2)
- SciTePress (2)
- Cornell University (1)
- De Gruyter Oldenbourg (1)
- Elsevier (1)
- Hometrica Consulting (1)
”I have never seen one who loves virtue as much as he loves beauty,” Confucius once said. If beauty is more important as goodness, it becomes clear why people invest so much effort in their first impression. The aesthetic of faces has many aspects and there is a strong correlation to all characteristics of humans, like age and gender. Often, research on aesthetics by social and ethic scientists lacks sufficient labelled data and the support of machine vision tools. In this position paper we propose the Aesthetic-Faces dataset, containing training data which is labelled by Chinese and German annotators. As a combination of three image subsets, the AF-dataset consists of European, Asian and African people. The research communities in machine learning, aesthetics and social ethics can benefit from our dataset and our toolbox. The toolbox provides many functions for machine learning with state-of-the-art CNNs and an Extreme-Gradient-Boosting regressor, but also 3D Morphable Model technolo gies for face shape evaluation and we discuss how to train an aesthetic estimator considering culture and ethics.
The aim of this work is the development of artificial intelligence (AI) application to support the recruiting process that elevates the domain of human resource management by advancing its capabilities and effectiveness. This affects recruiting processes and includes solutions for active sourcing, i.e. active recruitment, pre-sorting, evaluating structured video interviews and discovering internal training potential. This work highlights four novel approaches to ethical machine learning. The first is precise machine learning for ethically relevant properties in image recognition, which focuses on accurately detecting and analysing these properties. The second is the detection of bias in training data, allowing for the identification and removal of distortions that could skew results. The third is minimising bias, which involves actively working to reduce bias in machine learning models. Finally, an unsupervised architecture is introduced that can learn fair results even without ground truth data. Together, these approaches represent important steps forward in creating ethical and unbiased machine learning systems.
Most Question-answering (QA) systems rely on training data to reach their optimal performance. However, acquiring training data for supervised systems is both time-consuming and resource-intensive. To address this, in this paper, we propose TFCSG, an unsupervised similar question retrieval approach that leverages pre-trained language models and multi-task learning. Firstly, topic keywords in question sentences are extracted sequentially based on a latent topic-filtering algorithm to construct unsupervised training corpus data. Then, the multi-task learning method is used to build the question retrieval model. There are three tasks designed. The first is a short sentence contrastive learning task. The second is the question sentence and its corresponding topic sequence similarity judgment task. The third is using question sentences to generate their corresponding topic sequence task. The three tasks are used to train the language model in parallel. Finally, similar questions are obtained by calculating the cosine similarity between sentence vectors. The comparison experiment on public question datasets that TFCSG outperforms the comparative unsupervised baseline method. And there is no need for manual marking, which greatly saves human resources.
3D assisted 2D face recognition involves the process of reconstructing 3D faces from 2D images and solving the problem of face recognition in 3D. To facilitate the use of deep neural networks, a 3D face, normally represented as a 3D mesh of vertices and its corresponding surface texture, is remapped to image-like square isomaps by a conformal mapping. Based on previous work, we assume that face recognition benefits more from texture. In this work, we focus on the surface texture and its discriminatory information content for recognition purposes. Our approach is to prepare a 3D mesh, the corresponding surface texture and the original 2D image as triple input for the recognition network, to show that 3D data is useful for face recognition. Texture enhancement methods to control the texture fusion process are introduced and we adapt data augmentation methods. Our results show that texture-map-based face recognition can not only compete with state-of-the-art systems under the same precon ditions but also outperforms standard 2D methods from recent years.
In this paper, it aims to model wind speed time series at multiple sites. The five-parameter Johnson distribution is deployed to relate the wind speed at each site to a Gaussian time series, and the resultant m-dimensional Gaussian stochastic vector process Z(t) is employed to model the temporal-spatial correlation of wind speeds at m different sites. In general, it is computationally tedious to obtain the autocorrelation functions (ACFs) and cross-correlation functions (CCFs) of Z(t), which are different to those of wind speed times series. In order to circumvent this correlation distortion problem, the rank ACF and rank CCF are introduced to characterize the temporal-spatial correlation of wind speeds, whereby the ACFs and CCFs of Z(t) can be analytically obtained. Then, Fourier transformation is implemented to establish the cross-spectral density matrix of Z(t), and an analytical approach is proposed to generate samples of wind speeds at m different sites. Finally, simulation experiments are performed to check the proposed methods, and the results verify that the five-parameter Johnson distribution can accurately match distribution functions of wind speeds, and the spectral representation method can well reproduce the temporal-spatial correlation of wind speeds.
For collision and obstacle avoidance as well as trajectory planning, robots usually generate and use a simple 2D costmap without any semantic information about the detected obstacles. Thus a robot’s path planning will simply adhere to an arbitrarily large safety margin around obstacles. A more optimal approach is to adjust this safety margin according to the class of an obstacle. For class prediction, an image processing convolutional neural network can be trained. One of the problems in the development and training of any neural network is the creation of a training dataset. The first part of this work describes methods and free open source software, allowing a fast generation of annotated datasets. Our pipeline can be applied to various objects and environment settings and is extremely easy to use to anyone for synthesising training data from 3D source data. We create a fully synthetic industrial environment dataset with 10 k physically-based rendered images and annotations. Our da taset and sources are publicly available at https://github.com/LJMP/synthetic-industrial-dataset. Subsequently, we train a convolutional neural network with our dataset for costmap safety class prediction. We analyse different class combinations and show that learning the safety classes end-to-end directly with a small dataset, instead of using a class lookup table, improves the quantity and precision of the predictions.
We present a fully automatic approach to real-time 3D face reconstruction from monocular in-the-wild videos. We use a 3D morphable face model to obtain a semi-dense shape and combine it with a fast median-based super-resolution technique to obtain a high-fidelity textured 3D face model. Our system does not need prior training and is designed to work in uncontrolled scenarios.
We present a fully automatic approach to real-time 3D face reconstruction from monocular in-the-wild videos. With the use of a cascaded-regressor-based face tracking and a 3D morphable face model shape fitting, we obtain a semidense 3D face shape. We further use the texture information from multiple frames to build a holistic 3D face representation from the video footage. Our system is able to capture facial expressions and does not require any person specific training. We demonstrate the robustness of our approach on the challenging 300 Videos in the Wild (300- VW) dataset. Our real-time fitting framework is available as an open-source library at http://4dface.org.
With the rapid development of globalization, the demand for translation between different languages is also increasing. Although pre-training has achieved excellent results in neural machine translation, the existing neural machine translation has almost no high-quality suitable for specific fields. Alignment information, so this paper proposes a pre-training neural machine translation with alignment information via optimal transport. First, this paper narrows the representation gap between different languages by using OTAP to generate domain-specific data for information alignment, and learns richer semantic information. Secondly, this paper proposes a lightweight model DR-Reformer, which uses Reformer as the backbone network, adds Dropout layers and Reduction layers, reduces model parameters without losing accuracy, and improves computational efficiency. Experiments on the Chinese and English datasets of AI Challenger 2018 and WMT-17 show that the proposed algorithm has better performance than existing algorithms.
In clothing e-commerce, the challenge of optimally recommending clothing that suits a user’s unique characteristics remains a pressing issue. Many platforms simply recommend best-selling or popular clothing, without taking into account important attributes like user’s face color, pupil color, face shape, age, etc. To solve this problem, this paper proposes a personalized clothing recommendation algorithm that incorporates the established 4-Season Color System and user-specific biological characteristics. Firstly, the attributes and colors of clothing are classified by Fnet network, that can learn disjoint label combinations and mitigate the issue of excessive labels. Secondly, on the basis of the 4-Season Color System, the user’s face color model is trained by combined MobileNetV3_DTL, which ensures the model’s generalization and improves the training speed. Thirdly, user’s face shape and age are divided into different categories by an Inception network. Finally, according to the users’ face color, age, face shape and other information, personalized clothing is recommended in a coarse-to-fine manner. Experiments on five datasets demonstrate that the algorithm proposed in this paper achieves state-of-the-art results.
With the continuous development of economy, consumers pay more attention to the demand for personalization clothing. However, the recommendation quality of the existing clothing recommendation system is not enough to meet the user’s needs. When browsing online clothing, facial expression is the salient information to understand the user’s preference. In this paper, we propose a novel method to automatically personalize clothing recommendation based on user emotional analysis. Firstly, the facial expression is classified by multiclass SVM. Next, the user’s multi-interest value is calculated using expression intensity that is obtained by hybrid RCNN. Finally, the multi-interest value is fused to carry out personalized recommendation. The experimental results show that the proposed method achieves a significant improvement over other algorithms.
We presented our robot framework and our efforts to make face analysis more robust towards self-occlusion caused by head pose. By using a lightweight linear fitting algorithm, we are able to obtain 3D models of human faces in real-time. The combination of adaptive tracking and 3D face modelling for the analysis of human faces is used as a basis for further research on human-machine interaction on our SCITOS robot platform.
Socially interactive robots with human-like speech synthesis and recognition, coupled with humanoid appearance, are an important subject of robotics and artificial intelligence research. Modern solutions have matured enough to provide simple services to human users. To make the interaction with them as fast and intuitive as possible, researchers strive to create transparent interfaces close to human-human interaction. Because facial expressions play a central role in human-human communication, robot faces were implemented with varying degrees of human-likeness and expressiveness. We propose a way to implement a program that believably animates changing facial expressions and allows to influence them via inter-process communication based on an emotion model. This will can be used to create a screen based virtual face for a robotic system with an inviting appearance to stimulate users to seek interaction with the robot.
AI-based prediction and recommender systems are widely used in various industry sectors. However, general acceptance of AI-enabled systems is still widely uninvestigated. Therefore, firstly we conducted a survey with 559 respondents. Findings suggested that AI-enabled systems should be fair, transparent, consider personality traits and perform tasks efficiently. Secondly, we developed a system for the Facial Beauty Prediction (FBP) benchmark that automatically evaluates facial attractiveness. As our previous experiments have proven, these results are usually highly correlated with human ratings. Consequently they also reflect human bias in annotations. An upcoming challenge for scientists is to provide training data and AI algorithms that can withstand distorted information. In this work, we introduce AntiDiscriminationNet (ADN), a superior attractiveness prediction network. We propose a new method to generate an unbiased convolutional neural network (CNN) to improve the fairn ess of machine learning in facial dataset. To train unbiased networks we generate synthetic images and weight training data for anti-discrimination assessments towards different ethnicities. Additionally, we introduce an approach with entropy penalty terms to reduce the bias of our CNN. Our research provides insights in how to train and build fair machine learning models for facial image analysis by minimising implicit biases. Our AntiDiscriminationNet finally outperforms all competitors in the FBP benchmark by achieving a Pearson correlation coefficient of PCC = 0.9601.
This paper presents a novel multi-modal CNN architecture that exploits complementary input cues in addition to sole color information. The joint model implements a mid-level fusion that allows the network to exploit cross modal interdependencies already on a medium feature-level. The benefit of the presented architecture is shown for the RGB-D image understanding task. So far, state-of-the-art RGB-D CNNs have used network weights trained on color data. In contrast, a superior initialization scheme is proposed to pre-train the depth branch of the multi-modal CNN independently. In an end-to-end training the network parameters are optimized jointly using the challenging Cityscapes dataset. In thorough experiments, the effectiveness of the proposed model is shown. Both, the RGB GoogLeNet and further RGB-D baselines are outperformed with a significant margin on two different tasks: semantic segmentation and object detection. For the latter, this paper shows how to extract object level groundtruth from the instance level annotations in Cityscapes in order to train a powerful object detector.
Deep learning-based fabric defect detection methods have been widely investigated to improve production efficiency and product quality. Although deep learning-based methods have proved to be powerful tools for classification and segmentation, some key issues remain to be addressed when applied to real applications. Firstly, the actual fabric production conditions of factories necessitate higher real-time performance of methods. Moreover, fabric defects as abnormal samples are very rare compared with normal samples, which results in data imbalance. It makes model training based on deep learning challenging. To solve these problems, an extremely efficient convolutional neural network, Mobile-Unet, is proposed to achieve the end-to-end defect segmentation. The median frequency balancing loss function is used to overcome the challenge of sample imbalance. Additionally, Mobile-Unet introduces depth-wise separable convolution, which dramatically reduces the complexity cost and model size of the network. It comprises two parts: encoder and decoder. The MobileNetV2 feature extractor is used as the encoder, and then five deconvolution layers are added as the decoder. Finally, the softmax layer is used to generate the segmentation mask. The performance of the proposed model has been evaluated by public fabric datasets and self-built fabric datasets. In comparison with other methods, the experimental results demonstrate that segmentation accuracy and detection speed in the proposed method achieve state-of-the-art performance.
SLAM systems are mainly applied for robot navigation while research on feasibility for motion planning with SLAM for tasks like bin-picking, is scarce. Accurate 3D reconstruction of objects and environments is important for planning motion and computing optimal gripper pose to grasp objects. In this work, we propose the methods to analyze the accuracy of a 3D environment reconstructed using a LSD-SLAM system with a monocular camera mounted onto the gripper of a collaborative robot. We discuss and propose a solution to the pose space conversion problem. Finally, we present several criteria to analyze the 3D reconstruction accuracy. These could be used as guidelines to improve the accuracy of 3D reconstructions with monocular LSD-SLAM and other SLAM based solutions.
Understanding the factors that influence the accuracy of visual SLAM algorithms is very important for the future development of these algorithms. So far very few studies have done this. In this paper, a simulation model is presented and used to investigate the effect of the number of scene points tracked, the effect of the baseline length in triangulation and the influence of image point location uncertainty. It is shown that the latter is very critical, while the other all play important roles. Experiments with a well known semi-dense visual SLAM approach are also presented, when used in a monocular visual odometry mode. The experiments show that not including sensor bias and scale factor uncertainty is very detrimental to the accuracy of the simulation results.
In recent years, 3D facial reconstructions from single images have garnered significant interest. Most of the approaches are based on 3D Morphable Model (3DMM) fitting to reconstruct the 3D face shape. Concurrently, the adoption of Generative Adversarial Networks (GAN) has been gaining momentum to improve the texture of reconstructed faces. In this paper, we propose a fundamentally different approach to reconstructing the 3D head shape from a single image by harnessing the power of GAN. Our method predicts three maps of normal vectors of the head’s frontal, left, and right poses. We are thus presenting a model-free method that does not require any prior knowledge of the object’s geometry to be reconstructed.
The key advantage of our proposed approach is the substantial improvement in reconstruction quality compared to existing methods, particularly in the case of facial regions that are self-occluded in the input image. Our method is not limited to 3d face reconstruction. It is generic and applicable to multiple kinds of 3D objects. To illustrate the versatility of our method, we demonstrate its efficacy in reconstructing the entire human body.
By delivering a model-free method capable of generating high-quality 3D reconstructions, this paper not only advances the field of 3D facial reconstruction but also provides a foundation for future research and applications spanning multiple object types. The implications of this work have the potential to extend far beyond facial reconstruction, paving the way for innovative solutions and discoveries in various domains.
In visual adaptive tracking, the tracker adapts to the target, background, and conditions of the image sequence. Each update introduces some error, so the tracker might drift away from the target over time. To increase the robustness against the drifting problem, we present three ideas on top of a particle filter framework: An optical-flow-based motion estimation, a learning strategy for preventing bad updates while staying adaptive, and a sliding window detector for failure detection and finding the best training examples. We experimentally evaluate the ideas using the BoBoT dataseta. The code of our tracker is available online.