Nvidia Corporation

University of California

Orazio Gallo

Jan Kautz

Kari Pulli

Jinwei Gu

Jun Hu

Ming-Yu Liu

Dawid Stanislaw Pajak

Felix Heide

Nagilla Dikpal Reddy

David Edward Jacobs

Mushfiqur Rouf

Patrick Wieschollek

Kihwan Kim

Alejandro Jose Troccoli

Seung-Hwan Baek

Roberto Manduchi

Abhishek Haridas Badki

Huaijin Chen

Hang Su

Zitian Chen

Apparatuses, systems, and techniques to identify objects in view of a camera associated with a vehicle. In at least one embodiment, objects with which a vehicle may collide are identified, based on, for example, a difference between a size of an image of the objects detected at a first point in time and a size of an image of the objects detected at a subsequent point in time.

Object detection and collision avoidance using a neural network

Apparatuses, systems, and techniques to identify object distance with one or more cameras. In at least one embodiment, one or more cameras capture at least two images, where one image is transformed to the other, and a neural network determines whether said object is in front of or behind a known distance, whereby an object's distance may be determined after a set of known distances are analyzed.

Depth estimation using a neural network

One embodiment of a method includes predicting one or more three-dimensional (3D) mesh representations based on a plurality of digital images, wherein the one or more 3D mesh representations are refined by minimizing at least one difference between the one or more 3D mesh representations and the plurality of digital images.

Mesh reconstruction using data-driven priors

A target image corresponding to a novel view may be synthesized from two source images, corresponding source camera poses, and pixel attribute correspondences between the two source images. A particular object in the target image need only be visible in one of the two source images for successful synthesis. Each pixel in the target image is defined according to an identified pixel in one of the two source images. The identified source pixel provides attributes such as color, texture, and feature descriptors for the target pixel. The source and target camera poses are used to define geometric relationships for identifying the source pixels. In an embodiment, the pixel attribute correspondences are optical flow that defines movement of attributes from a first image of the two source images to a second image of the two source images.

View synthesis using attribute correspondences and geometric relationship constraints

When a computer image is generated from a real-world scene having a semi-reflective surface (e.g. window), the computer image will create, at the semi-reflective surface from the viewpoint of the camera, both a reflection of a scene in front of the semi-reflective surface and a transmission of a scene located behind the semi-reflective surface. Similar to a person viewing the real-world scene from different locations, angles, etc., the reflection and transmission may change, and also move relative to each other, as the viewpoint of the camera changes. Unfortunately, the dynamic nature of the reflection and transmission negatively impacts the performance of many computer applications, but performance can generally be improved if the reflection and transmission are separated. The present disclosure uses deep learning to separate reflection and transmission at a semi-reflective surface of a computer image generated from a real-world scene.

Deep-learning method for separating reflection and transmission images visible at a semi-reflective surface in a computer image of a real-world scene

Missing image content is generated using a neural network. In an embodiment, a high resolution image and associated high resolution semantic label map are generated from a low resolution image and associated low resolution semantic label map. The input image/map pair (low resolution image and associated low resolution semantic label map) lacks detail and is therefore missing content. Rather than simply enhancing the input image/map pair, data missing in the input image/map pair is improvised or hallucinated by a neural network, creating plausible content while maintaining spatio-temporal consistency. Missing content is hallucinated to generate a detailed zoomed in portion of an image. Missing content is hallucinated to generate different variations of an image, such as different seasons or weather conditions for a driving video.

Guided hallucination for missing image content using a neural network

An image processing method extracts consecutive input blurry frames from a video, and generates sharp frames corresponding to the input blurry frames. An optical flow is determined between the sharp frames, and the optical flow is used to compute a per-pixel blur kernel. The blur kernel is used to reblur each of the sharp frames into a corresponding re-blurred frame. The re-blurred frame is used to fine-tune the deblur network by minimizing the distance between the re-blurred frame and the input blurry frame.

Unsupervised learning approach for video deblurring

A system and method for computational zoom generates a resulting image having two or more effective focal lengths. A first surface within a three-dimensional (3D) scene including a first and second set of 3D objects defined by 3D information is identified. The first and second sets of 3D objects are located within first and second depth ranges of the 3D scene, respectively. The first set of 3D objects is projected onto the first surface according to a first projection mapping to produce a first portion of image components. The second set of 3D objects is projected onto the first surface according to a second projection mapping to produce a second portion of image components. The resulting image comprising the first portion of image components and the second portion of image components is generated based on a camera projection from the first surface to a camera view plane.

Image generation of a three-dimensional scene using multiple focal lengths

A computer implemented method of determining a latent image from an observed image is disclosed. The method comprises implementing a plurality of image processing operations within a single optimization framework, wherein the single optimization framework comprises solving a linear minimization expression. The method further comprises mapping the linear minimization expression onto at least one non-linear solver. Further, the method comprises using the non-linear solver, iteratively solving the linear minimization expression in order to extract the latent image from the observed image, wherein the linear minimization expression comprises: a data term, and a regularization term, and wherein the regularization term comprises a plurality of non-linear image priors.

Unified optimization method for end-to-end camera image processing for translating a sensor captured image to a display image

A system, computer-readable medium, and method are provided for generating images based on adaptations of the human visual system. An input image is received, an effect provoking change is received, and an afterimage resulting from a cumulative effect of human visual adaptation is computed based on the effect provoking change and a per-photoreceptor type physiological adaptation of the human visual system. The computed afterimage may include a bleaching afterimage effect and/or a local adaptation afterimage effect. The computed afterimage is then accumulated into an output image for display.

Physiologically based adaptive image generation

A number of images of a scene are captured and stored. The images are captured over a range of values for an attribute (e.g., a camera setting). One of the images is displayed. A location of interest in the displayed image is identified. Regions that correspond to the location of interest are identified in each of the images. Those regions are evaluated to identify which of the regions is rated highest with respect to the attribute relative to the other regions. The image that includes the highest-rated region is then displayed.

Interaction with and display of photographic images in an image stack

A system, method, and computer program product are provided for performing fast, non-rigid registration for at least two images of a high-dynamic range image stack. The method includes the steps of generating a warped image based on a set of corresponding pixels, analyzing the warped image to detect unreliable pixels in the warped image, and generating a corrected pixel value for each unreliable pixel in the warped image. The set of corresponding pixels includes a plurality of pixels in a source image, each pixel in the plurality of pixels associated with a potential feature in the source image and paired with a corresponding pixel in a reference image that substantially matches the pixel in the source image.

System, method, and computer program product for performing fast, non-rigid registration for high dynamic range image stacks

Barcode decoding bypassing binarization is provided which relies on deformable templates to makes use of all the gray level information of each pixel in the barcode image. Parameterization of the deformable templates allow for efficiently performing maximum likelihood estimation independently on each barcode digit and enforcing spatial coherence across the barcode digits.

Growing community of inventors

Santa Cruz, CA, United States of America

Orazio Gallo