Disney Enterprises, Inc.

Eidgenossische Technische Hochschule Zurich

Gaspard Zoss

Derek Edward Bradley

Paulo Fabiano Urnau Gotardo

Prashanth Chandran

Dominik Thabo Beeler

Eftychios Sifakis

Jeremy Riviere

Sebastian Winberg

Yingyan Xu

Aurel Gruber

Lingchen Yang

Daoye Wang

Simone Foti

Byungsoo Kim

Baran Gözcü

Marco Fratarcangeli

Barbara Solenthaler

A technique for generating a sequence of geometries includes converting, via an encoder neural network, one or more input geometries corresponding to one or more frames within an animation into one or more latent vectors. The technique also includes generating the sequence of geometries corresponding to a sequence of frames within the animation based on the one or more latent vectors. The technique further includes causing output related to the animation to be generated based on the sequence of geometries.

Synthesizing sequences of 3D geometries for movement-based performance

Techniques are disclosed for generating photorealistic images of head portraits. A rendering application renders a set of images that include the skin of a face and corresponding masks indicating pixels associated with the skin in the images. An inpainting application performs a neural projection technique to optimize a set of parameters that, when input into a generator model, produces a set of projection images, each of which includes a head portrait in which (1) skin regions resemble the skin regions of the face in a corresponding rendered image; and (2) non-skin regions match the non-skin regions in the other projection images when the rendered set of images are standalone images, or transition smoothly between consecutive projection images in the case when the rendered set of images are frames of a video. The rendered images can then be blended with corresponding projection images to generate composite images that are photorealistic.

Techniques for enhancing skin renders using neural network projection for rendering completion

Methods and systems for generating three-dimensional (3D) models and facial hair models representative of subjects (e.g., actors or actresses) using facial scanning technology. Initial subject facial data, including facial frames and facial performance frames (e.g., images of the subject collected from a capture system) can be used to accurately predict the structure of the subject's face underneath their facial hair to produce a reference 3D facial shape of the subject. Likewise, image processing techniques can be used to identify facial hairs and generate a reference facial hair model. The reference 3D facial shape and reference facial hair mode can subsequently be used to generate performance 3D facial shapes and a performance facial hair model corresponding to a performance by the subject (e.g., reciting dialog).

Generating a facial-hair-free mesh of a subject

Embodiments of the present disclosure are directed to methods and systems for generating three-dimensional (3D) models and facial hair models representative of subjects (e.g., actors or actresses) using facial scanning technology. Methods accord to embodiments may be useful for performing facial capture on subjects with dense facial hair. Initial subject facial data, including facial frames and facial performance frames (e.g., images of the subject collected from a capture system) can be used to accurately predict the structure of the subject's face underneath their facial hair to produce a reference 3D facial shape of the subject. Likewise, image processing techniques can be used to identify facial hairs and generate a reference facial hair model. The reference 3D facial shape and reference facial hair mode can subsequently be used to generate performance 3D facial shapes and a performance facial hair model corresponding to a performance by the subject (e.g., reciting dialog).

Dynamic facial hair capture of a subject

Various embodiments include a system for rendering an object, such as human skin or a human head, from captured appearance data. The system includes a processor executing a near field lighting reconstruction module. The system determines at least one of a three-dimensional (3D) position or a 3D orientation of a lighting unit based on a plurality of captured images of a mirror sphere. For each point light source in a plurality of point light sources included in the lighting unit, the system determines an intensity associated with the point light source. The system determines captures appearance data of the object, where the object is illuminated by the lighting unit. The system renders an image of the object based on the appearance data and the intensities associated with each point light source in the plurality of point light sources.

Techniques for improved lighting models for appearance capture

A technique for performing style transfer between a content sample and a style sample is disclosed. The technique includes applying one or more neural network layers to a first latent representation of the style sample to generate one or more convolutional kernels. The technique also includes generating convolutional output by convolving a second latent representation of the content sample with the one or more convolutional kernels. The technique further includes applying one or more decoder layers to the convolutional output to produce a style transfer result that comprises one or more content-based attributes of the content sample and one or more style-based attributes of the style sample.

Adaptive convolutions in neural networks

Various embodiments include a system for rendering an object, such as human skin or a human head, from captured appearance data comprising a plurality of texels. The system includes a processor executing a texture space indirect illumination module. The system determines texture coordinates of a vector originating from a first texel where the vector intersects a second texel. The system renders the second texel from the viewpoint of the first texel based on appearance data at the second texel. Based on the rendering of the second texel, the system determines an indirect lighting intensity incident to the first texel from the second texel. The system updates appearance data at the first texel based on a direct lighting intensity and the indirect lighting intensity. The system renders the first texel based on the updated appearance data at the first texel.

Techniques are disclosed for re-aging images of faces and three-dimensional (3D) geometry representing faces. In some embodiments, an image of a face, an input age, and a target age, are input into a re-aging model, which outputs a re-aging delta image that can be combined with the input image to generate a re-aged image of the face. In some embodiments, 3D geometry representing a face is re-aged using local 3D re-aging models that each include a blendshape model for finding a linear combination of sample patches from geometries of different facial identities and generating a new shape for the patch at a target age based on the linear combination. In some embodiments, 3D geometry representing a face is re-aged by performing a shape-from-shading technique using re-aged images of the face captured from different viewpoints, which can optionally be constrained to linear combinations of sample patches from local blendshape models.

Techniques for re-aging faces in images and video frames

Embodiment of the present invention sets forth techniques for performing face reconstruction. The techniques include generating an identity mesh based on an identity encoding that represents an identity associated with a face in one or more images. The techniques also include generating an expression mesh based on an expression encoding that represents an expression associated with the face in the one or more images. The techniques also include generating, by a machine learning model, an output mesh of the face based on the identity mesh and the expression mesh.

Face reconstruction using a mesh convolution network

A technique for rendering an input geometry includes generating a first segmentation mask for a first input geometry and a first set of texture maps associated with one or more portions of the first input geometry. The technique also includes generating, via one or more neural networks, a first set of neural textures for the one or more portions of the first input geometry. The technique further includes rendering a first image corresponding to the first input geometry based on the first segmentation mask, the first set of texture maps, and the first set of neural textures.

Synthesizing sequences of images for movement-based performance

Techniques are disclosed for generating photorealistic images of objects, such as heads, from multiple viewpoints. In some embodiments, a morphable radiance field (MoRF) model that generates images of heads includes an identity model that maps an identifier (ID) code associated with a head into two codes: a deformation ID code encoding a geometric deformation from a canonical head geometry, and a canonical ID code encoding a canonical appearance within a shape-normalized space. The MoRF model also includes a deformation field model that maps a world space position to a shape-normalized space position based on the deformation ID code. Further, the MoRF model includes a canonical neural radiance field (NeRF) model that includes a density multi-layer perceptron (MLP) branch, a diffuse MLP branch, and a specular MLP branch that output densities, diffuse colors, and specular colors, respectively. The MoRF model can be used to render images of heads from various viewpoints.

Techniques for multi-view neural object modeling

One embodiment of the present invention sets forth a technique for generating actuation values based on a target shape such that the actuation values cause a simulator to output a simulated soft body that matches the target shape. The technique includes inputting a latent code that represents a target shape and a point on a geometric mesh into a first machine learning model. The technique further includes generating, via execution of the first machine learning model, one or more simulator control values that specify a deformation of the geometric mesh, where each of the simulator control values is based on the latent code and corresponds to the input point, and generating, via execution of the simulator, a simulated soft body based on the one or more simulator control values and the geometric mesh. The technique further includes causing the simulated soft body to be outputted to a computing device.

Data-driven physics-based models with implicit actuations

A technique for synthesizing a shape includes generating a first plurality of offset tokens based on a first shape code and a first plurality of position tokens, wherein the first shape code represents a variation of a canonical shape, and wherein the first plurality of position tokens represent a first plurality of positions on the canonical shape. The technique also includes generating a first plurality of offsets associated with the first plurality of positions on the canonical shape based on the first plurality of offset tokens. The technique further includes generating the shape based on the first plurality of offsets and the first plurality of positions.

Transformer-based shape models

Some implementations of the disclosure are directed to capturing facial training data for one or more subjects, the captured facial training data including each of the one or more subject's facial skin geometry tracked over a plurality of times and the subject's corresponding jaw poses for each of those plurality of times; and using the captured facial training data to create a model that provides a mapping from skin motion to jaw motion. Additional implementations of the disclosure are directed to determining a facial skin geometry of a subject; using a model that provides a mapping from skin motion to jaw motion to predict a motion of the subject's jaw from a rest pose given the facial skin geometry; and determining a jaw pose of the subject using the predicted motion of the subject's jaw.

Jaw tracking without markers for facial performance capture

A modeling engine generates a prediction model that quantifies and predicts secondary dynamics associated with the face of a performer enacting a performance. The modeling engine generates a set of geometric representations that represents the face of the performer enacting different facial expressions under a range of loading conditions. For a given facial expression and specific loading condition, the modeling engine trains a Machine Learning model to predict how soft tissue regions of the face of the performer change in response to external forces applied to the performer during the performance. The modeling engine combines different expression models associated with different facial expressions to generate a prediction model. The prediction model can be used to predict and remove secondary dynamics from a given geometric representation of a performance or to generate and add secondary dynamics to a given geometric representation of a performance.

Growing community of inventors

Zurich, Switzerland

Gaspard Zoss