Abstract: Systems and techniques are provided for deep learning model optimizations. An example process can include generating, based on processing data using a neural network model, a plurality of prediction outputs associated with a plurality of output channels corresponding to a multi-channel prediction target; determining that a prediction output from the plurality of prediction outputs has a value that is outside of a quantization range and one or more remaining prediction outputs from the plurality of prediction outputs have a respective value that is within the quantization range; clamping the prediction output based on the quantization range; and generating a single channel output based on the clamped prediction output and the one or more remaining prediction outputs.

Abstract: Systems and techniques are provided for generating a texture for a three-dimensional (3D) facial model. For example, a process can include obtaining a first frame, the first frame including a first portion of a face. In some aspects, the process can include generating a 3D facial model based on the first frame and generating a first facial feature corresponding to the first portion of the face. In some examples, the process includes obtaining a second frame, the second frame including a second portion of the face. In some cases, the second portion of the face at least partially overlaps the first portion of the face. In some examples, the process includes combining the first facial feature with the second facial feature to generate an enhanced facial feature, wherein the combining is performed to enhance an appearance of select areas of the enhanced facial feature.

Abstract: Systems and techniques are provided for generating a three-dimensional (3D) facial model. For example, a process can include obtaining at least one input image associated with a face. In some aspects, the process can include obtaining a pose for a 3D facial model associated with the face. In some examples, the process can include generating, by a machine learning model, the 3D facial model associated with the face. In some cases, one or more parameters associated with a shape component of the 3D facial model are conditioned on the pose. In some implementations, the 3D facial model is configured to vary in shape based on the pose for the 3D facial model associated with the face.

Abstract: Methods, systems, and apparatuses are provided to automatically reconstruct an image, such as a 3D image. For example, a computing device may obtain an image, and may apply a first trained machine learning process to the image to generate coefficient values characterizing the image in a plurality of dimensions. Further, the computing device may generate a mesh based on the coefficient values. The computing device may apply a second trained machine learning process to the coefficient values and the image to generate a displacement map. Based on the mesh and the displacement map, the computing device may generate output data characterizing an aligned mesh. The computing device may store the output data within a data repository. In some examples, the computing device provides the output data for display.

Abstract: Techniques are provided for generating one or more three-dimensional (3D) models. In one example, an image of an object (e.g., a face or other object) is obtained, and a 3D model of the object in the image is generated. The 3D model includes geometry information. Color information for the 3D model is determined, and a fitted 3D model of the object is generated based on a modification of the geometry information and the color information for the 3D model. In some cases, the color information (e.g., determination and/or modification of the color information) and the fitted 3D model can be based on one or more vertex-level fitting processes. A refined 3D model of the object is generated based on the fitted 3D model and depth information associated with the fitted 3D model. In some cases, the refined 3D model can be based on a pixel-level refinement or fitting process.

Abstract: Systems and techniques are provided for performing video-based activity recognition. For example, a process can include generating a three-dimensional (3D) model of a first portion of an object based on one or more frames depicting the object. The process can also include generating a mask for the one or more frames, the mask including an indication of one or more regions of the object. The process can further include generating a 3D base model based on the 3D model of the first portion of the object and the mask, the 3D base model representing the first portion of the object and a second portion of the object. The process can include generating, based on the mask and the 3D base model, a 3D model of the second portion of the object.

Abstract: Techniques are provided for generating one or more three-dimensional (3D) models. In one example, an image of an object (e.g., a face or other object) is obtained, and a 3D model of the object in the image is generated. The 3D model includes geometry information. Color information for the 3D model is determined, and a fitted 3D model of the object is generated based on a modification of the geometry information and the color information for the 3D model. In some cases, the color information (e.g., determination and/or modification of the color information) and the fitted 3D model can be based on one or more vertex-level fitting processes. A refined 3D model of the object is generated based on the fitted 3D model and depth information associated with the fitted 3D model. In some cases, the refined 3D model can be based on a pixel-level refinement or fitting process.