Recent breakthroughs in synthetic data generation approaches made it possible to produce highly photorealistic images which are hardly distinguishable from real ones. Furthermore, synthetic generation pipelines have the potential to generate an unlimited number of images. The combination of high photorealism and scale turn synthetic data into a promising candidate for improving various machine learning (ML) pipelines. Thus far, a large body of research in this field has focused on using synthetic images for training, by augmenting and enlarging training data. In contrast to using synthetic data for training, in this work we explore whether synthetic data can be beneficial for model selection. Considering the task of image classification, we demonstrate that when data is scarce, synthetic data can be used to replace the held out validation set, thus allowing to train on a larger dataset. We also introduce a novel method to calibrate the synthetic error estimation to fit that of the real domain. We show that such calibration significantly improves the usefulness of synthetic data for model selection.
RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection
College of Information and Engineering, Capital Normal University(首都师范大学信息工程学院);School of Artificial Intelligence, Beijing University of Posts and Telecommunications(北京邮电大学人工智能学院)
Synthetic Sample Selection via Reinforcement Learning
College of Information Sciences and Technology, The Pennsylvania State University(宾夕法尼亚州立大学信息科学与技术学院);National Library of Medicine, National Institutes of Health(美国国立卫生研究院国家医学图书馆);College of Medicine, The Pennsylvania State University(宾夕法尼亚州立大学医学院)
Representation and Reference Selection in Training-Free Synthetic Image Attribution
College of Computer Science and Artificial Intelligence, Fudan University(复旦大学计算机科学与人工智能学院);Department of Information Engineering and Mathematics, University of Siena(锡耶纳大学信息工程与数学系)
Grasp Pre-shape Selection by Synthetic Training: Eye-in-hand Shared Control on the Hannes Prosthesis
Humanoid Sensing and Perception, Istituto Italiano di Tecnologia(人机感知与感知,意大利技术研究所);Rehab Technologies Lab, Istituto Italiano di Tecnologia(康复技术实验室,意大利技术研究所);DIBRIS, University of Genoa(Genoa大学DIBRIS)
Visual Anomaly Synthesis for Model Selection in Data Scarcity
Digital Factory Vorarlberg GmbH(福拉尔贝格数字工厂有限公司);illwerke vkw AG(伊尔沃克VKW股份公司);University of Salzburg(萨尔茨堡大学);Vorarlberg University of Applied Sciences(福拉尔贝格应用科学大学)