Abstract: A method includes obtaining a first image of a scene using a first image sensor of an electronic device and a second image of the scene using a second image sensor of the electronic device. The method also includes generating a first feature map from the first image and a second feature map from the second image. The method further includes generating a third feature map based on the first feature map, the second feature map, and an asymmetric search window. The method additionally includes generating a depth map by restoring spatial resolution to the third feature map.

Abstract: A method to map a plurality of feature maps of a neural network onto a memory hierarchy includes mapping a first feature map of the plurality of feature maps to a memory in a memory hierarchy having available memory space and providing quickest access to the first feature map. The method also includes, when the first feature map expires, removing the first feature map from the memory used to store the first feature map.

Abstract: A method includes obtaining, using at least one processor, first and second input image frames, where the first and second input image frames are associated with first and second image planes, respectively. The method also includes obtaining, using the at least one processor, a depth map associated with the first input image frame. The method further includes producing another version of the depth map by performing one or more times: (a) projecting, using the at least one processor, the first input image frame to the second image plane in order to produce a projected image frame using (i) the depth map and (ii) information identifying a conversion from the first image plane to the second image plane and (b) adjusting, using the at least one processor, at least one of the depth map and the information identifying the conversion from the first image plane to the second image plane.

Abstract: A method includes processing, using at least one processor of an electronic device, each of multiple images using a photometric augmentation engine, where the photometric augmentation engine performs one or more photometric augmentation operations. The method also includes applying, using the at least one processor, multiple layers of a convolutional neural network to each of the images, where each layer generates a corresponding feature map. The method further includes processing, using the at least one processor, at least one of the feature maps using at least one feature augmentation engine between consecutive layers of the multiple layers, where the at least one feature augmentation engine performs one or more feature augmentation operations.

Abstract: A method includes obtaining at least three input image frames of a scene captured using at least three imaging sensors. The input image frames include a reference image frame and multiple non-reference image frames. The method also includes generating multiple disparity maps using the input image frames. Each disparity map is associated with the reference image frame and a different non-reference image frame. The method further includes generating multiple confidence maps using the input image frames. Each confidence map identifies weights associated with one of the disparity maps. In addition, the method includes generating a depth map of the scene using the disparity maps and the confidence maps. The imaging sensors are arranged to define multiple baseline directions, where each baseline direction extends between the imaging sensor used to capture the reference image frame and the imaging sensor used to capture a different non-reference image frame.

Abstract: A method for performing a fundamental computational primitive in a device is provided, where the device includes a processor and a matrix multiplication accelerator (MMA). The method includes configuring a streaming engine in the device to stream data for the fundamental computational primitive from memory, configuring the MMA to format the data, and executing the fundamental computational primitive by the device.

Abstract: A method includes obtaining, using first and second image sensors of an electronic device, first and second images, respectively, of a scene. The method also includes obtaining, using an image depth sensor of the electronic device, a third image and a first depth map of the scene, the first depth map having a resolution lower than a resolution of the first and second images. The method further includes undistorting the first and second images using the third image and the first depth map. The method also includes rectifying the first and second images using the third image and the first depth map. The method further includes generating a disparity map using the first and second images that have been undistorted and rectified. In addition, the method includes generating a second depth map using the disparity map and the first depth map, where the second depth map has a resolution that is higher than the resolution of the first depth map.

Abstract: Methods and systems for compressed domain progressive application of computer vision techniques. A method for decoding video data includes receiving a video stream that is encoded for multi-stage decoding. The method includes partially decoding the video stream by performing one or more stages of the multi-stage decoding. The method includes determining whether a decision for a computer vision system can be identified based on the partially decoded video stream. Additionally, the method includes generating the decision for the computer vision system based on decoding of the video stream. A system for encoding video data includes a processor configured to receive the video data from a camera, encode the video data received from the camera into a video stream for consumption by a computer vision system, and include metadata with the encoded video stream to indicate whether a decision for the computer vision system can be identified from the metadata.

Abstract: A method includes identifying, by at least one processor, multiple features of input data using a common feature extractor. The method also includes processing, by the at least one processor, at least some identified features using each of multiple pre-processing branches. Each pre-processing branch includes a first set of neural network layers and generates initial outputs associated with a different one of multiple data processing tasks. The method further includes combining, by the at least one processor, at least two initial outputs from at least two pre-processing branches to produce combined initial outputs. In addition, the method includes processing, by the at least one processor, at least some initial outputs or at least some combined initial outputs using each of multiple post-processing branches. Each post-processing branch includes a second set of neural network layers and generates final outputs associated with a different one of the multiple data processing tasks.

Abstract: A method of capturing a photo using a mobile device is described. The method comprises detecting a signature motion of the mobile device; capturing video using a camera of the mobile device upon detecting the signature motion; and automatically identifying a start frame and an end frame of the captured images using neural networks. A mobile device for capturing a photo is also described.

Abstract: A method includes obtaining a first image of a scene using a first image sensor of an electronic device and a second image of the scene using a second image sensor of the electronic device. The method also includes generating a first feature map from the first image and a second feature map from the second image. The method further includes generating a third feature map based on the first feature map, the second feature map, and an asymmetric search window. The method additionally includes generating a depth map by restoring spatial resolution to the third feature map.

Abstract: An electronic device, method, and computer readable medium for 3D association of detected objects are provided. The electronic device includes a memory and at least one processor coupled to the memory. The at least one processor configured to convolve an input to a neural network with a basis kernel to generate a convolution result, scale the convolution result by a scalar to create a scaled convolution result, and combine the scaled convolution result with one or more of a plurality of scaled convolution results to generate an output feature map.

Abstract: An electronic device, method, and computer readable medium for an invertible wavelet layer for neural networks are provided. The electronic device includes a memory and at least one processor coupled to the memory. The at least one processor is configured to receive an input to a neural network, apply a wavelet transform to the input at a wavelet layer of the neural network, and generate a plurality of subbands of the input as a result of the wavelet transform.

Abstract: A method, an electronic device, and computer readable medium are provided. The method includes receiving an input into a neural network that includes a kernel. The method also includes generating, during a convolution operation of the neural network, multiple panel matrices based on different portions of the input. The method additionally includes successively combining each of the multiple panel matrices with the kernel to generate an output. Generating the multiple panel matrices can include mapping elements within a moving window of the input onto columns of an indexing matrix, where a size of the window corresponds to the size of the kernel.

Abstract: A method of capturing a photo using a mobile device is described. The method comprises detecting a signature motion of the mobile device; capturing video using a camera of the mobile device upon detecting the signature motion; and automatically identifying a start frame and an end frame of the captured images using neural networks. A mobile device for capturing a photo is also described.

Abstract: Methods and systems for compressed domain progressive application of computer vision techniques. A method for decoding video data includes receiving a video stream that is encoded for multi-stage decoding. The method includes partially decoding the video stream by performing one or more stages of the multi-stage decoding. The method includes determining whether a decision for a computer vision system can be identified based on the partially decoded video stream. Additionally, the method includes generating the decision for the computer vision system based on decoding of the video stream. A system for encoding video data includes a processor configured to receive the video data from a camera, encode the video data received from the camera into a video stream for consumption by a computer vision system, and include metadata with the encoded video stream to indicate whether a decision for the computer vision system can be identified from the metadata.

Abstract: A method to map a plurality of feature maps of a neural network onto a memory hierarchy includes mapping a first feature map of the plurality of feature maps to a memory in a memory hierarchy having available memory space and providing quickest access to the first feature map. The method also includes, when the first feature map expires, removing the first feature map from the memory used to store the first feature map.

Abstract: A method for performing a fundamental computational primitive in a device is provided, where the device includes a processor and a matrix multiplication accelerator (MMA). The method includes configuring a streaming engine in the device to stream data for the fundamental computational primitive from memory, configuring the MMA to format the data, and executing the fundamental computational primitive by the device.

Abstract: For protecting a speaker, an input signal is received, and an excursion of the speaker that would be caused by the input signal is predicted. In response to the predicted excursion exceeding a threshold, a targeted excursion of the speaker is determined by compressing the predicted excursion. The targeted excursion is translated into an output signal, which is output to the speaker.

Abstract: A capacitive touch keypad may receive a touch by a conductive object to one or more keys of the keypad, in which the keys each have a separate capacitive key electrode. The touched locations may be determined by precharging a pseudo-randomly selected set of key electrodes a number (M) of iterations while the keyboard is being touched, producing a sense voltage for each iteration by coupling all of the key electrodes to a reference capacitor, producing a 1-bit sense measurement for each iteration by comparing the sense voltage to a reference voltage to form a set of M 1-bit sense measurement, and by applying a modified binary iterative hard threshold (BIHT) algorithm to the set of M 1-bit sense measurements, in which only the non-zero locations of the solution are monitored in each iteration.