Refining Linked Data with Games with a Purpose

With the rise of linked data and knowledge graphs,the need becomes compelling to find suitable solutions to increase the coverage and correctness of data sets,to add missing knowledge and to identify and remove errors.Several approaches-mostly relying on machine learning and natural language processing techniques-have been proposed to address this refinement goal;they usually need a partial gold standard,i.e.,some“ground truth”to train automatic models.Gold standards are manually constructed,either by involving domain experts or by adopting crowdsourcing and human computation solutions.In this paper,we present an open source software framework to build Games with a Purpose for linked data refinement,i.e.,Web applications to crowdsource partial ground truth,by motivating user participation through fun incentive.We detail the impact of this new resource by explaining the specific data linking“purposes”supported by the framework(creation,ranking and validation of links)and by defining the respective crowdsourcing tasks to achieve those goals.We also introduce our approach for incremental truth inference over the contributions provided by players of Games with a Purpose(also abbreviated as GWAP):we motivate the need for such a method with the specificity of GWAP vs.traditional crowdsourcing;we explain and formalize the proposed process,explain its positive consequences and illustrate the results of an experimental comparison with state-of-the-art approaches.To show this resource’s versatility,we describe a set of diverse applications that we built on top of it;to demonstrate its reusability and extensibility potential,we provide references to detailed documentation,including an entire tutorial which in a few hours guides new adopters to customize and adapt the framework to a new use case.

Discriminative feature encoding for intrinsic image decomposition

Intrinsic image decomposition is an important and long-standing computer vision problem.Given an input image,recovering the physical scene properties is ill-posed.Several physically motivated priors have been used to restrict the solution space of the optimization problem for intrinsic image decomposition.This work takes advantage of deep learning,and shows that it can solve this challenging computer vision problem with high efficiency.The focus lies in the feature encoding phase to extract discriminative features for different intrinsic layers from an input image.To achieve this goal,we explore the distinctive characteristics of different intrinsic components in the high-dimensional feature embedding space.We define feature distribution divergence to efficiently separate the feature vectors of different intrinsic components.The feature distributions are also constrained to fit the real ones through a feature distribution consistency.In addition,a data refinement approach is provided to remove data inconsistency from the Sintel dataset,making it more suitable for intrinsic image decomposition.Our method is also extended to intrinsic video decomposition based on pixel-wise correspondences between adjacent frames.Experimental results indicate that our proposed network structure can outperform the existing state-of-the-art.