Few-shot object detection: a comprehensive survey. - In: IEEE transactions on neural networks and learning systems, ISSN 2162-2388, Bd. 0 (2023), 0, S. 1-21
Humans are able to learn to recognize new objects even from a few examples. In contrast, training deep-learning-based object detectors requires huge amounts of annotated data. To avoid the need to acquire and annotate these huge amounts of data, few-shot object detection (FSOD) aims to learn from few object instances of new categories in the target domain. In this survey, we provide an overview of the state of the art in FSOD. We categorize approaches according to their training scheme and architectural layout. For each type of approach, we describe the general realization as well as concepts to improve the performance on novel categories. Whenever appropriate, we give short takeaways regarding these concepts in order to highlight the best ideas. Eventually, we introduce commonly used datasets and their evaluation protocols and analyze the reported benchmark results. As a result, we emphasize common challenges in evaluation and identify the most promising current trends in this emerging field of FSOD.
https://doi.org/10.1109/TNNLS.2023.3265051
Analysis of appeal for realistic AI-generated photos. - In: IEEE access, ISSN 2169-3536, Bd. 11 (2023), S. 38999-39012
AI-generated images have gained in popularity in recent years due to improvements and developments in the field of artificial intelligence. This has led to several new AI generators, which may produce realistic, funny, and impressive images using a simple text prompt. DALL-E-2, Midjourney, and Craiyon are a few examples of the mentioned approaches. In general, it can be seen that the quality, realism, and appeal of the images vary depending on the used approach. Therefore, in this paper, we analyze to what extent such AI-generated images are realistic or of high appeal from a more photographic point of view and how users perceive them. To evaluate the appeal of several state-of-the-art AI generators, we develop a dataset consisting of 27 different text prompts, with some of them being based on the DrawBench prompts. Using these prompts we generated a total of 135 images with five different AI-Text-To-Image generators. These images in combination with real photos form the basis of our evaluation. The evaluation is based on an online subjective study and the results are compared with state-of-the-art image quality models and features. The results indicate that some of the included generators are able to produce realistic and highly appealing images. However, this depends on the approach and text prompt to a large extent. The dataset and evaluation of this paper are made publicly available for reproducibility, following an Open Science approach.
https://doi.org/10.1109/ACCESS.2023.3267968
Influence of the temperature-time regime on the mechanical properties during the DED-Arc process of near-net-shape Ti-6Al-4 V components. - In: Welding in the world, ISSN 1878-6669, Bd. 67 (2023), 7, S. 1643-1665
In a research project, the additive manufacturing process of components made of Ti-6Al-4 V using gas metal arc welding (GMAW), which is classified into the directed energy deposition-arc (DED-Arc) processes, was investigated. The project focused on the systematic development of economical additive build-up strategies and the analysis of the temperature-time regime during the build-up process, as well as the investigation of the resulting properties. A welding range diagram was created with recommendations for process settings for additive manufacturing with the controlled short circuit, as well as a presentation of possible defect patterns outside the range shown. For the fabrication of thick-walled structures, various build-up strategies were investigated by modifying the welding path and evaluated with regard to their suitability. Based on the results, additive structures were fabricated by varying the temperature-time regime in order to gain insights into selected geometrical, metallurgical, and mechanical properties. Different energy inputs per unit length, structure dimensions, and interpass temperatures (IPT) were used for this purpose. The research project provides comprehensive findings on the additive processing of the material Ti-6Al-4 V using metal inert gas welding, in particular with regard to the temperature-time regime and the resulting properties.
https://doi.org/10.1007/s40194-023-01513-7
Interactive robot teaching based on finger trajectory using multimodal RGB-D-T-data. - In: Frontiers in robotics and AI, ISSN 2296-9144, Bd. 10 (2023), 1120357, S. 01-13
The concept of Industry 4.0 brings the change of industry manufacturing patterns that become more efficient and more flexible. In response to this tendency, an efficient robot teaching approach without complex programming has become a popular research direction. Therefore, we propose an interactive finger-touch based robot teaching schema using a multimodal 3D image (color (RGB), thermal (T) and point cloud (3D)) processing. Here, the resulting heat trace touching the object surface will be analyzed on multimodal data, in order to precisely identify the true hand/object contact points. These identified contact points are used to calculate the robot path directly. To optimize the identification of the contact points we propose a calculation scheme using a number of anchor points which are first predicted by hand/object point cloud segmentation. Subsequently a probability density function is defined to calculate the prior probability distribution of true finger trace. The temperature in the neighborhood of each anchor point is then dynamically analyzed to calculate the likelihood. Experiments show that the trajectories estimated by our multimodal method have significantly better accuracy and smoothness than only by analyzing point cloud and static temperature distribution.
https://doi.org/10.3389/frobt.2023.1120357
Model predictive control of parabolic PDE systems under chance constraints. - In: Mathematics, ISSN 2227-7390, Bd. 11 (2023), 6, 1372, S. 1-23
Model predictive control (MPC) heavily relies on the accuracy of the system model. Nevertheless, process models naturally contain random parameters. To derive a reliable solution, it is necessary to design a stochastic MPC. This work studies the chance constrained MPC of systems described by parabolic partial differential equations (PDEs) with random parameters. Inequality constraints on time- and space-dependent state variables are defined in terms of chance constraints. Using a discretization scheme, the resulting high-dimensional chance constrained optimization problem is solved by our recently developed inner-outer approximation which renders the problem computationally amenable. The proposed MPC scheme automatically generates probability tubes significantly simplifying the derivation of feasible solutions. We demonstrate the viability and versatility of the approach through a case study of tumor hyperthermia cancer treatment control, where the randomness arises from the thermal conductivity coefficient characterizing heat flux in human tissue.
https://doi.org/10.3390/math11061372
Structural and optical properties of gold nanosponges revealed via 3D nano-reconstruction and phase-field models. - In: Communications materials, ISSN 2662-4443, Bd. 4 (2023), 1, 20, S. 1-13
Nanosponges are subject of intensive research due to their unique morphology, which leads among other effects to electrodynamic field localization generating a strongly nonlinear optical response at hot spots and thus enable a variety of applications. Accurate predictions of physical properties require detailed knowledge of the sponges’ chaotic nanometer-sized structure, posing a metrological challenge. A major goal is to obtain computer models with equivalent structural and optical properties. Here, to understand the sponges’ morphology, we present a procedure for their accurate 3D reconstruction using focused ion beam tomography. Additionally, we introduce a simulation method to create nanoporous sponge models with adjustable geometric properties. It is shown that if certain morphological parameters are similar for computer-generated and experimental sponges, their optical response, including magnitudes and hot spot locations, are also similar. Finally, we analyze the anisotropy of experimental sponges and present an easy-to-use method to reproduce arbitrary anisotropies in computer-generated sponges.
https://doi.org/10.1038/s43246-023-00346-7
Modeling of an automatic vision mixer with human characteristics for multi-camera theater recordings. - In: IEEE access, ISSN 2169-3536, Bd. 11 (2023), S. 18714-18726
A production process using high-resolution cameras can be used for multi-camera recordings of theater performances or other stage performances. One approach to automate the generation of suitable image cuts could be to focus on speaker changes so that the person who is speaking is shown in the generated cut. However, these image cuts can appear static and robotic if they are set too precisely. Therefore, the characteristics and habits of professional vision mixers (persons who operate the vision mixing desk) during the editing process are investigated in more detail in order to incorporate them into an automation process. The characteristic features of five different vision mixers are examined, which were used under almost identical recording conditions for theatrical cuts in TV productions. The cuts are examined with regard to their temporal position in relation to pauses in speech, which take place during speaker changes on stage. It is shown that different professional vision mixers set the cuts individually differently before, in or after the pauses in speech. Measured are differences on average up to 0.3 seconds. From the analysis of the image cuts, an approach for a model is developed in which the individual characteristics of a vision mixer can be set. With the help of this novel model, a more human appearance can be given to otherwise exact and robotic cuts, when automating image cuts.
https://doi.org/10.1109/ACCESS.2023.3245804
The HAInich: a multidisciplinary vision data-set for a better understanding of the forest ecosystem. - In: Scientific data, ISSN 2052-4463, Bd. 10 (2023), 1, 168, S. 1-11
We present a multidisciplinary forest ecosystem 3D perception dataset. The dataset was collected in the Hainich-Dün region in central Germany, which includes two dedicated areas, which are part of the Biodiversity Exploratories - a long term research platform for comparative and experimental biodiversity and ecosystem research. The dataset combines several disciplines, including computer science and robotics, biology, bio-geochemistry, and forestry science. We present results for common 3D perception tasks, including classification, depth estimation, localization, and path planning. We combine the full suite of modern perception sensors, including high-resolution fisheye cameras, 3D dense LiDAR, differential GPS, and an inertial measurement unit, with ecological metadata of the area, including stand age, diameter, exact 3D position, and species. The dataset consists of three hand held measurement series taken from sensors mounted on a UAV during each of three seasons: winter, spring, and early summer. This enables new research opportunities and paves the way for testing forest environment 3D perception tasks and mission set automation for robotics.
https://doi.org/10.1038/s41597-023-02010-8
SenGlove - a modular wearable device to measure kinematic parameters of the human hand. - In: Bioengineering, ISSN 2306-5354, Bd. 10 (2023), 3, 324, S. 1-29
For technical or medical applications, the knowledge of the exact kinematics of the human hand is key to utilizing its capability of handling and manipulating objects and communicating with other humans or machines. The optimal relationship between the number of measurement parameters, measurement accuracy, as well as complexity, usability and cost of the measuring systems is hard to find. Biomechanic assumptions, the concepts of a biomechatronic system and the mechatronic design process, as well as commercially available components, are used to develop a sensorized glove. The proposed wearable introduced in this paper can measure 14 of 15 angular values of a simplified hand model. Additionally, five contact pressure values at the fingertips and inertial data of the whole hand with six degrees of freedom are gathered. Due to the modular design and a hand size examination based on anthropometric parameters, the concept of the wearable is applicable to a large variety of hand sizes and adaptable to different use cases. Validations show a combined root-mean-square error of 0.99° to 2.38° for the measurement of all joint angles on one finger, surpassing the human perception threshold and the current state-of-the-art in science and technology for comparable systems.
https://doi.org/10.3390/bioengineering10030324
Ultrahigh-rate and ultralong-duration sodium storage enabled by sodiation-driven reconfiguration. - In: Advanced energy materials, ISSN 1614-6840, Bd. 13 (2023), 6, 2204324, S. 1-12
Despite their variable valence and favorable sodiation/desodiation potential, vanadium sulfides (VSx) used as anode materials of sodium-ion batteries (SIBs) have been held back by their capacity decline and low cycling capability, associated with the structure distortion volume expansion and pulverization. This study reports an accessible process to tackle these challenges via fabricating a 3D-VSx anode for SIBs with ultrahigh-rate and ultralong-duration stable sodium storage. The sodiation-driven reactivation of micro-nano 3D-VSx activates the reconfiguration effect, effectively maintaining structural integrity. Interestingly, the mechanical degradation of 3D-VSx over the sodiation process can be controlled by fine-tuning the operating voltage. The self-reconfigured open nanostructures with large void space not only effectively withstand repetitive volume changes and mitigate the damaging mechanical stresses, but also in turn construct a self-optimized shortened ion diffusion pathway. Moreover, the sodiation-driven reconfiguration excites many active sites and optimizes a stable solid-electrolyte interface, thereby delivering a reversible capacity of 961.4 mA h g^-1 after 1500 cycles at a high rate of 2 A g^-1. This work provides new insight into the rational design of electrodes toward long-lived SIBs through sodiation-driven reconfiguration.
https://doi.org/10.1002/aenm.202204324