Abstract: A computer-implemented method for computing a prediction on images of a scene includes: receiving one or more polarization raw frames of a scene, the polarization raw frames being captured with a polarizing filter at a different linear polarization angle; extracting one or more first tensors in one or more polarization representation spaces from the polarization raw frames; and computing a prediction regarding one or more optically challenging objects in the scene based on the one or more first tensors in the one or more polarization representation spaces.

Abstract: One embodiment of the present application sets forth a wearable device that includes a display comprising a plurality of pixels and configured to emit light, and a micro-lens array located adjacent to the display, and configured to produce a light field by altering the light emitted by the display, where at least one of the display or the micro-lens array is configured to move from a first position to a second position that aligns a first pixel in the plurality of pixels relative to the micro-lens array.

Abstract: A system for facilitating stress-adaptive virtual experience station includes a virtual display system and camera system coupled to a kinematic rig. An array of transducers coupled to the station interact with a user based on a feedback signal, configured for user health treatments, audio effects, or computational imaging techniques.

Abstract: A multi-modal sensor system includes: an underlying sensor system; a polarization camera system configured to capture polarization raw frames corresponding to a plurality of different polarization states; and a processing system including a processor and memory, the processing system being configured to control the underlying sensor system and the polarization camera system, the memory storing instructions that, when executed by the processor, cause the processor to: control the underlying sensor system to perform sensing on a scene and the polarization camera system to capture a plurality of polarization raw frames of the scene; extract first tensors in polarization representation spaces based on the plurality of polarization raw frames; and compute a characterization output based on an output of the underlying sensor system and the first tensors in polarization representation spaces.

Abstract: A deep neural network may be trained on the data of one or more entities, also know as Alices. An outside computing entity, also known as a Bob, may assist in these computations, without receiving access to Alices' data. Data privacy may be preserved by employing a “split” neural network. The network may comprise an Alice part and a Bob part. The Alice part may comprise at least three neural layers, and the Bob part may comprise at least two neural layers. When training on data of an Alice, that Alice may input her data into the Alice part, perform forward propagation though the Alice part, and then pass output activations for the final layer of the Alice part to Bob. Bob may then forward propagate through the Bob part. Similarly, backpropagation may proceed backwards through the Bob part, and then through the Alice part of the network.

Abstract: A light source may illuminate a scene that is obscured by fog. Light may reflect back to a time-resolved light sensor. For instance, the light sensor may comprise avalanche photodiodes that are not single-photon sensitive. The light sensor may perform a raster scan. The imaging system may determine reflectance and depth of the fog-obscured target. The imaging system may perform a probabilistic algorithm that exploits the fact that times of arrival of photons reflected from fog have a Gamma distribution that is different than the Gaussian distribution of times of arrival of photons reflected from the target. The imaging system may adjust frame rate locally depending on local density of fog, as indicated by a local Gamma distribution determined in a prior step. The imaging system may perform one or more of spatial regularization, temporal regularization, and deblurring.

Abstract: A system for collecting data for training a computer vision model for shape estimation includes: an imaging system configured to capture one or more images; and a processing system including a processor and memory storing instructions that, when executed by the processor, cause the processor to: receive one or more input images from the imaging system; estimate a pose of an object depicted in the one or more images; render a shape estimate from a 3-D model of the object posed in accordance with the pose of the object; and generate a data point of a training dataset, the data point including one or more images based on the one or more input images and a label corresponding to the one or more images, the label including the shape estimate.

Abstract: A multi-modal sensor system includes: an underlying sensor system; a polarization camera system configured to capture polarization raw frames corresponding to a plurality of different polarization states; and a processing system including a processor and memory, the processing system being configured to control the underlying sensor system and the polarization camera system, the memory storing instructions that, when executed by the processor, cause the processor to: control the underlying sensor system to perform sensing on a scene and the polarization camera system to capture a plurality of polarization raw frames of the scene; extract first tensors in polarization representation spaces based on the plurality of polarization raw frames; and compute a characterization output based on an output of the underlying sensor system and the first tensors in polarization representation spaces.

Abstract: A distributed deep learning network may prevent an attacker from reconstructing raw data from activation outputs of an intermediate layer of the network. To achieve this, the loss function of the network may tend to reduce distance correlation between raw data and the activation outputs. For instance, the loss function may be the sum of two terms, where the first term is weighted distance correlation between raw data and activation outputs of a split layer of the network, and the second term is weighted categorical cross entropy of actual labels and label predictions. Distance correlation with the entire raw data may be minimized. Alternatively, distance correlation with only with certain features of the raw data may be minimized, in order to ensure attribute-level privacy. In some cases, a client computer calculates decorrelated representations of raw data before sharing information about the data with external computers.

Abstract: A computer-implemented method for computing a prediction on images of a scene includes: receiving one or more polarization raw frames of a scene, the polarization raw frames being captured with a polarizing filter at a different linear polarization angle; extracting one or more first tensors in one or more polarization representation spaces from the polarization raw frames; and computing a prediction regarding one or more optically challenging objects in the scene based on the one or more first tensors in the one or more polarization representation spaces.

Abstract: A computer-implemented method for computing a prediction on images of a scene includes: receiving one or more polarization raw frames of a scene, the polarization raw frames being captured with a polarizing filter at a different linear polarization angle; extracting one or more first tensors in one or more polarization representation spaces from the polarization raw frames; and computing a prediction regarding one or more optically challenging objects in the scene based on the one or more first tensors in the one or more polarization representation spaces.

Abstract: A time-of-flight imaging system may output light with a modulation frequency in the gigahertz band, to illuminate a range target. This high-frequency illumination may enable extremely precise—e.g., micron-scale—depth measurements. The system may modulate reflected light from the range target, to create a beat tone that has a frequency in the hertz band. In some cases, the modulated light in the gigahertz band is created by a first modulator and the beat tone in the hertz band is created by a second modulator. In some cases, the modulated light in the gigahertz band is created by an upshift cascade of modulators and the beat tone in the hertz band is created by a downshift cascade of modulators. A photodetector may measure the low-frequency beat tone. From this beat tone, phase of the signal and depth of the range target may be extracted.

Abstract: A multi-modal sensor system includes: an underlying sensor system; a polarization camera system configured to capture polarization raw frames corresponding to a plurality of different polarization states; and a processing system including a processor and memory, the processing system being configured to control the underlying sensor system and the polarization camera system, the memory storing instructions that, when executed by the processor, cause the processor to: control the underlying sensor system to perform sensing on a scene and the polarization camera system to capture a plurality of polarization raw frames of the scene; extract first tensors in polarization representation spaces based on the plurality of polarization raw frames; and compute a characterization output based on an output of the underlying sensor system and the first tensors in polarization representation spaces.

Abstract: One embodiment of the present application sets forth a wearable device that includes a display comprising a plurality of pixels and configured to emit light, and a micro-lens array located adjacent to the display, and configured to produce a light field by altering the light emitted by the display, where at least one of the display or the micro-lens array is configured to move from a first position to a second position that aligns a first pixel in the plurality of pixels relative to the micro-lens array.

Abstract: A computer-implemented method for computing a prediction on images of a scene includes: receiving one or more polarization raw frames of a scene, the polarization raw frames being captured with a polarizing filter at a different linear polarization angle; extracting one or more first tensors in one or more polarization representation spaces from the polarization raw frames; and computing a prediction regarding one or more optically challenging objects in the scene based on the one or more first tensors in the one or more polarization representation spaces.

Abstract: In one embodiment, a method includes receiving data associated with the user from a content rendering device associated with a user of an online social network, where the content rendering device receives signals from a broadcast service provider system, identifying personalized content items that are of interest to the user based at least on the received data associated with the user, sending the identified personalized content items to the content rendering device, where the identified personalized content items are used by the content rendering device to determine which content items to cache, ranking content objects in the online social network based at least on the identified personalized content items, sending one or more content objects that have higher ranks than the other content objects to the broadcast service provider system, where the one or more content objects are sent to the content rendering device by the broadcast service provider system.

Abstract: A sample may be illuminated in such a way that light passes through the sample, reflects from a set of reflectors, passes through the sample again and travels to a light sensor. The reflectors may be staggered in depth beneath the sample, each reflector being at a different depth. Light reflecting from each reflector, respectively, may arrive at the light sensor during a different time interval than that in which light reflecting from other reflectors arrives—or may have a different phase than that of light reflecting from the other reflectors. The light sensor may separately measure light reflecting from each reflector, respectively. The reflectors may be extremely small, and the separate reflections from the different reflectors may be combined in a super-resolved image. The super-resolved image may have a spatial resolution that is better than that indicated by the diffraction limit.

Abstract: An otoscope may project a temporal sequence of phase-shifted fringe patterns onto an eardrum, while a camera in the otoscope captures images. A computer may calculate a global component of these images. Based on this global component, the computer may output an image of the middle ear and eardrum. This image may show middle ear structures, such as the stapes and incus. Thus, the otoscope may “see through” the eardrum to visualize the middle ear. The otoscope may project another temporal sequence of phase-shifted fringe patterns onto the eardrum, while the camera captures additional images. The computer may subtract a fraction of the global component from each of these additional images. Based on the resulting direct-component images, the computer may calculate a 3D map of the eardrum.

Abstract: In one embodiment, a computing system may determine that a performance metric of an eye tracking system is below a performance threshold. The eye tracking system may be associated with a head-mounted display worn by a user. In response to the determination that the performance metric is below the performance threshold, the system may identify one or more contents being displayed by the head-mounted display. The system may access one or more properties associated with the one or more contents. The system may predict a vergence distance of the user based at least on the one or more properties associated with the one or more display contents. The system may adjust one or more configurations based on the predicted vergence distance of the user.

Abstract: A distributed deep learning network may prevent an attacker from reconstructing raw data from activation outputs of an intermediate layer of the network. To achieve this, the loss function of the network may tend to reduce distance correlation between raw data and the activation outputs. For instance, the loss function may be the sum of two terms, where the first term is weighted distance correlation between raw data and activation outputs of a split layer of the network, and the second term is weighted categorical cross entropy of actual labels and label predictions. Distance correlation with the entire raw data may be minimized. Alternatively, distance correlation with only with certain features of the raw data may be minimized, in order to ensure attribute-level privacy. In some cases, a client computer calculates decorrelated representations of raw data before sharing information about the data with external computers.