汤姆逊

Employees are a company’s most valuable assets. However, if they left their jobs suddenly, it might cost a company a lot of money. Consequently, companies nowadays are actively seeking tools and technologies that can help in accurate and early employee attrition prediction. The main focus here shall be on the visualization of the available employee data in order to gain intuitive insights on the correlation among attributes and top causes behind the attrition. In this paper, we present the comparison of four algorithms used to predict whether an employee will leave or not, based on various attributes like age, salary, experience, etc., along with intuitive visualizations and its importance. The models used for prediction are random forest, logistic regression, K-nearest neighbor, and Naïve Bayes classifier. Thus, the visual analysis of employee attrition problem and accurate prediction can allow HR managers to take precautionary actions to retain the employee within the company.

The training of deep neural network classifiers results in decision boundaries which geometry is still not well understood. This is in direct relation with classification problems such as so called adversarial examples. We introduce zoNNscan, an index that is intended to inform on the boundary uncertainty (in terms of the presence of other classes) around one given input datapoint. It is based on confidence entropy, and is implemented through sampling in the multidimensional ball surrounding that input. We detail the zoNNscan index, give an algorithm for approximating it, and finally illustrate its benefits on four applications, including two important problems for the adoption of deep networks in critical systems: adversarial examples and corner case inputs. We highlight that zoNNscan exhibits significantly higher values than for standard inputs in those two problem classes.

3D reconstruction of insects from photographs is a challenging task as it requires to tackle several problems such as strong out-of-focus areas in macro-photography, thin structures (insect legs and hairs), flat-colored surfaces (insects shells), non-Lambertian (shells specularities) and even translucent surfaces (wings). In this work, we first present a new lens-based image registration technique for accurate multi-focus stacking suitable for 3D reconstruction purposes while other methods create in-focus images for viewing purpose only. We then evaluate and compare the classical Multi-View-Stereo (MVS) reconstruction pipeline for small and complex objects with recent deep learning-based reconstruction methods such as the Neural Radiance Fields (NeRF) and the Neural Sparse Voxel Fields (NSVF). We present an assessment of different sources of errors for the considered methods. The results are compared both quantitatively and qualitatively across the different methods. From this analysis we present a series of practical guidelines for addressing the common issues of the reconstruction of small objects under challenging conditions.

StyleGAN generates photorealistic portrait images of faces with eyes, teeth, hair and context (neck, shoulders, background), but lacks a rig-like control over semantic face parameters that are interpretable in 3D, such as face pose, expressions, and scene illumination. Three-dimensional morphable face models (3DMMs) on the other hand offer control over the semantic parameters, but lack photorealism when rendered and only model the face interior, not other parts of a portrait image (hair, mouth interior, background). We present the first method to provide a face rig-like control over a pretrained and fixed StyleGAN via a 3DMM. A new rigging network, RigNet is trained between the 3DMM's semantic parameters and StyleGAN's input. The network is trained in a self-supervised manner, without the need for manual annotations. At test time, our method generates portrait images with the photorealism of StyleGAN and provides explicit control over the 3D semantic parameters of the face.

In Visual Effects, the creation of realistic facial performances is still a challenge that the industry is trying to overcome. Blendshape deformation is used to reproduce the action of different groups of muscles, which produces realistic static results. However, this is not sufficient to generate believable and detailed facial performances of animated digital characters. To increase the realism of facial performances, it is possible to enhance standard facial rigs using physical simulation approaches. However, setting up a simulation rig and controlling material properties according to the performance is not an easy task and could take a lot of time and iterations to get it right. We present a workflow that allows us to generate an activation map for the fibres of a set of superficial patches we callpseudo-muscles. The pseudo-muscles are automatically identified using k-means to cluster the data from the blendshape targets in the animation rig and compute the direction of their contraction (direction of the pseudo-muscle fibres). We use an Extended Position-Based Dynamics solver to add physical simulation to the facial animation, controlling the behaviour of simulation through the activation map. We show the results achieved using the proposed solution on two digital humans and one fantastic cartoon character, demonstrating that the identified pseudo-muscles approximate facial anatomy and the simulation properties are properly controlled, increasing the realism while preserving the work of animators.