A system and method are disclosed herein for developing a machine perception model in the omnidirectional image domain. The system and method utilize the knowledge distillation process to transfer and adapt knowledge from the perspective projection image domain to the omnidirectional image domain. A teacher model is pre-trained to perform the machine perception task in the perspective projection image. A student model is trained by adapting the pre-existing knowledge of the teacher model from the perspective projection image domain to the omnidirectional image domain. By way of this training, the student model learns to perform the same machine perception task, except in the omnidirectional image domain, using limited or no suitably labeled training data in the omnidirectional image domain.

A wearable display device includes a first body, a first electrical connector and a connecting component. The first body has a display interface, and the first electrical connector is disposed on the first body. The connecting component is connected with the first body and adapted to fix different second bodies onto the first body. The first body and the second body transfer a signal through the first electrical connector.

Digitally acquiring digital media content using a computer model simulation of a real venue in which the digital content is to be shown. Frontally projecting content from a virtual center of the simulated venue, the real venue having a frontal screen and two side screens. Digitally capturing the frontally projected digital media content from a view that a real frontal projector and two real side projectors would have in real life in order to “bake in” a warped transformation of the frontally projected digital media content. Projecting the warped, transformed frontally projected content through a frontal digital projector and two side digital projectors in the real venue, thus completing an illusion of a “cinematic window” of the digital media content in the real venue. Controlling the digital projectors using a digital server to feed the three digitally captured, warped media streams synchronously to the front, left and right digital projectors.

An embodiment may include a display processor, memory to store a 2D frame corresponding to a projection from a 360 video, and a quality selector to select a quality factor for a block of the 2D frame based on quality information from neighboring blocks of the 2D frame, including blocks which are neighboring only in the 360 video space. The system may also include a range adjuster to adjust a search range for the 2D frame based on a search area of the 2D frame, a viewport manager to determine if a request for a viewport of the 2D frame extends beyond a first edge of the 2D frame and to fill the requested viewport with wrap-around image information, and/or a motion estimator to estimate motion information based on both color information and depth information. Other embodiments are disclosed and claimed.

A system and method are disclosed for exhibiting wide-screen motion pictures, including in multiplex-style theaters, with an immersive, picture-dominance effect not previously obtainable in such theaters. The invention uses a zero-gain or nominal-gain curved screen to accommodate a wide-screen presentation, and a digital projector capable of delivering 14 to 22 foot-lamberts of light to the screen. The invention can support any theatrical aspect ratio for presentation, including the widest in use, 2.76:1. To eliminate image distortion, the invention uses warping software to correct the image for the geometry of the auditorium and the shape of the screen. This correction is established for every aspect ratio that will be accommodated in the auditorium in which the invention is installed, and aspect ratios can be changed to display content with different aspect ratios during a program.

A vehicle camera system captures and transmits video to a user device, which includes a viewing device for playback of the captured video, such as virtual reality or augmented reality glasses. A rendering map is generated that indicates which pixels of the video frame (as identified by particular coordinates of the video frame) correspond to which coordinates of a virtual sphere in which a portion of the video frame is rendered for display. When a video frame is received, the rendering map is used to determine the texture values (e.g., colors) for coordinates in the virtual sphere, which is used to generate the display for the user. This technique reduces the rendering time when a user turns his or her head (e.g., while in virtual reality) and so it reduces motion and/or virtual reality sickness induced by the rendering lag.

Systems, apparatuses and methods may provide for technology that generates, by a first neural network, an initial set of model weights based on input data and iteratively generates, by a second neural network, an updated set of model weights based on residual data associated with the initial set of model weights and the input data. Additionally, the technology may output a geometric model of the input data based on the updated set of model weights. In one example, the first neural network and the second neural network reduce the dependence of the geometric model on the number of data points in the input data.

Disclosed are methods and apparatuses for image data encoding/decoding. A method for decoding a 360-degree image includes the steps of: receiving a bitstream obtained by encoding a 360-degree image; generating a prediction image by making reference to syntax information obtained from the received bitstream; adding the generated prediction image to a residual image obtained by dequantizing and inverse-transforming the bitstream, so as to obtain a decoded image; and reconstructing the decoded image into a 360-degree image according to a projection format. Therefore, the performance of image data compression can be improved.

Systems and methods described herein relate to training a machine-learning-based monocular depth estimator. One embodiment selects a virtual image in a virtual dataset, the virtual dataset including a plurality of computer-generated virtual images; generates, from the virtual image in accordance with virtual-camera intrinsics, a point cloud in three-dimensional space based on ground-truth depth information associated with the virtual image; reprojects the point cloud back to two-dimensional image space in accordance with real-world camera intrinsics to generate a transformed virtual image; and trains the machine-learning-based monocular depth estimator, at least in part, using the transformed virtual image.

A spherical harmonic is defined which is an operationally optimal small finite subset of the infinite number of spherical harmonics allowed to exist mathematically. The composition of the subset differs depending on its position on virtual hemisphere. The subsets are further divided into small spherical tesserae whose dimensions vary depending on the distance from the hemispherical center. The images of the outside visual scenes are projected on the flat surface of the webcam and from there are read and recalculated programmatically as if the images have been projected on the hemisphere. rotational invariants are then computed in the smallest tesserae using numerical integration, and then invariants from neighboring tesserae are added to compute the rotational invariant of their union. Every computed invariant is checked with the library and stored there if there is no match. The rotational invariants are solely used for visual recognition and classification and operational decision making.