Abstract: Training deep neural networks requires a large amount of labeled training data. Conventionally, labeled training data is generated by gathering real images that are manually labelled which is very time-consuming. Instead of manually labelling a training dataset, domain randomization technique is used generate training data that is automatically labeled. The generated training data may be used to train neural networks for object detection and segmentation (labelling) tasks. In an embodiment, the generated training data includes synthetic input images generated by rendering three-dimensional (3D) objects of interest in a 3D scene. In an embodiment, the generated training data includes synthetic input images generated by rendering 3D objects of interest on a 2D background image. The 3D objects of interest are objects that a neural network is trained to detect and/or label.

Abstract: Fabric Attached Memory (FAM) provides a pool of memory that can be accessed by one or more processors, such as a graphics processing unit(s) (GPU)(s), over a network fabric. In one instance, a technique is disclosed for using imperfect processors as memory controllers to allow memory, which is local to the imperfect processors, to be accessed by other processors as fabric attached memory. In another instance, memory address compaction is used within the fabric elements to fully utilize the available memory space.

Abstract: This disclosure relates to a receiver that includes a clock and data recovery loop and a phase offset loop. The clock and data recovery loop may be controlled by a sum of gradients for a plurality of data interleaves. The phase offset loop may be controlled by an accumulated differential gradient for each of the data interleaves.

Abstract: The present disclosure provides various approaches for smart area monitoring suitable for parking garages or other areas. These approaches may include ROI-based occupancy detection to determine whether particular parking spots are occupied by leveraging image data from image sensors, such as cameras. These approaches may also include multi-sensor object tracking using multiple sensors that are distributed across an area that leverage both image data and spatial information regarding the area, to provide precise object tracking across the sensors. Further approaches relate to various architectures and configurations for smart area monitoring systems, as well as visualization and processing techniques. For example, as opposed to presenting video of an area captured by cameras, 3D renderings may be generated and played from metadata extracted from sensors around the area.

Abstract: In various examples, optical flow estimate (OFE) quality is improved when employing a hint-based algorithm in multi-level hierarchical motion estimation by using different scan orders at different resolution levels. A scan of an image performed with a scan order may initially leverage OFEs from a previous scan of the image, where the previous scan was performed using a different scan order. The OFEs leveraged from the previous scan are more likely to be of high accuracy until sufficient spatial hints are available to the hint-based algorithm for the scan to reduce the impact of potentially lower quality OFEs resulting from the different scan order of the previous scan.

Abstract: In various examples, recordings of gameplay sessions are enhanced by the application of special effects to relatively high(er) and/or low(er) interest durations of the gameplay sessions. Durations of relatively high(er) or low(er) predicted interest in a gameplay session are identified, for instance, based upon level of activity engaged in by a gamer during a particular gameplay session duration. Once identified, different variations of video characteristic(s) are applied to at least a portion of the identified durations for implementation during playback. The recordings may be generated and/or played back in real-time with a live gameplay session, or after completion of the gameplay session. Further, video data of the recordings themselves may be modified to include the special effects and/or indications of the durations and/or variations may be included in metadata and used for playback.

Abstract: In various examples, a deep neural network (DNN) is trained—using image data alone—to accurately predict distances to objects, obstacles, and/or a detected free-space boundary. The DNN may be trained with ground truth data that is generated using sensor data representative of motion of an ego-vehicle and/or sensor data from any number of depth predicting sensors—such as, without limitation, RADAR sensors, LIDAR sensors, and/or SONAR sensors. The DNN may be trained using two or more loss functions each corresponding to a particular portion of the environment that depth is predicted for, such that—in deployment—more accurate depth estimates for objects, obstacles, and/or the detected free-space boundary are computed by the DNN. In some embodiments, a sampling algorithm may be used to sample depth values corresponding to an input resolution of the DNN from a predicted depth map of the DNN at an output resolution of the DNN.

Abstract: A network device configured to perform scalable, in-network computations is described. The network device is configured to process pull requests and/or push requests from a plurality of endpoints connected to the network. A collective communication primitive from a particular endpoint can be received at a network device. The collective communication primitive is associated with a multicast region of a shared global address space and is mapped to a plurality of participating endpoints. The network device is configured to perform an in-network computation based on information received from the participating endpoints before forwarding a response to the collective communication primitive back to one or more of the participating endpoints. The endpoints can inject pull requests (e.g., load commands) and/or push requests (e.g., store commands) into the network. A multicast capability enables tasks, such as a reduction operation, to be offloaded to hardware in the network device.

Abstract: A technique utilizing speculative execution and rollback for performing data parallel training of a neural network model is disclosed. Activations for a layer of the neural network model are normalized during a speculative normalization operation using estimated normalization parameters associated with a partial population of a set of training data allocated to a particular processor. Normalization parameters associated with the total population of the set of training data are generated by a distributed reduce operation in parallel with the speculative normalization operation. An optional rollback operation can revert the activations to a pre-normalization state if the estimated normalization parameters for the partial population are subsequently determined to be inaccurate compared to the normalization parameters for the population of the set of training data distributed across a plurality of processors.

Abstract: The disclosure is directed to a method of operating a game system where virtual machines (VM) supporting the game session can be reused by a second user after a first user ends their game session. In another aspect, the VM can be shut down if the number of VMs exceeds a target number of VMs or if an abnormality is detected in the VM. In another aspect, VMs can be instantiated in order to meet a target number of VMs. In another aspect, a software application is disclosed to execute the methods described herein. In another aspect, a game services system is disclosed that can operate a VM group, maintain a status parameter of the instantiated VMs, and service user requests for game sessions.

Abstract: An adder circuit provides a first operand input and a second operand input to an XNOR cell. The XNOR cell transforms these inputs to a propagate signal that is applied to an OAT cell to produce a carry out signal. A third OAT cell transforms a third operand input and the propagate signal into a sum output signal.

Abstract: A game-agnostic event detector can be used to automatically identify game events. Game-specific configuration data can be used to specify types of pre-processing to be performed on media for a game session, as well as types of detectors to be used to detect events for the game. Event data for detected events can be written to an event log in a form that is both human- and process-readable. The event data can be used for various purposes, such as to generate highlight videos or provide player performance feedback. The event data may be determined based upon output from detectors such as optical character recognition (OCR) engines, and the regions may be upscaled and binarized before OCR processing.

Abstract: A technique for selecting locations of tear lines when displaying visual content. The technique includes receiving coordinates for one or more portions of a display where a tear is permitted and determining if a frame transition is to occur while rendered content is being scanned out for display within the one or more portions of the display where tear is permitted. If the frame transition is to occur while the scanline for the display is in the one or more portions of the display where tear is permitted, then the technique further includes allowing the frame transition to occur. If the frame transition is to occur while the scanline for the display is not in the one or more portions of the display where tear is permitted, then the technique further includes delaying the frame transition until at least when the scanline for the display is in the one or more portions of the display where tear is permitted.

Abstract: One aspect of the disclosure provides a method for rendering an image. The method includes: placing primitives of the image in a screen space; binning the primitives into tiles of the screen space that the primitives touch; and rasterizing the tiles. The aforementioned rasterizing includes shading a subset of the primitives binned to one of the tiles over multiple passes at multiple shading rates, each of the shading rates is based at least on a frequency at which a color being shaded at each pass changes across the screen space, and the subset of the primitives are cached in an on-chip memory of a processor rendering the image between the passes.

Abstract: A switch architecture enables ports to stash packets in unused buffers on other ports, exploiting excess internal bandwidth that may exist, for example, in a tiled switch. This architecture leverages unused port buffer memory to improve features such as congestion handling and error recovery.

Abstract: A method dynamically selects one of a first sampling order and a second sampling order for a ray trace of pixels in a tile where the selection is based on a motion vector for the tile. The sampling order may be a bowtie pattern or an hourglass pattern.

Abstract: An augmented reality display system includes a first beam path for a foveal inset image on a holographic optical element, a second beam path for a peripheral display image on the holographic optical element, and pupil position tracking logic that generates control signals to set a position of the foveal inset as perceived through the holographic optical element, to determine the peripheral display image, and to control a moveable stage.

Abstract: A technique for evaluating qualitative streaming experience using session performance metadata is disclosed herein. A pipeline of a streaming service can be adapted to collect metadata, such as timestamps, from various components of the pipeline. The metadata can then be analyzed to calculate an objective quality metric for each streaming session using weighted scores derived from the metadata for a plurality of different components including, but not limited to, stutter, latency, and/or picture quality. The quality metric is designed to have high correlation with subjective measures of quality by users of the streaming service, but provides dense data samples compared to typical sparse responses collected from user feedback (e.g., user surveys). The objective quality metric can be utilized to quickly adjust, either manually or automatically, the streaming service parameters to improve the quality of the streaming service due to changes in, e.g., streaming content.

Abstract: A computer system comprising a processor and a memory for storing instructions, that when executed by the processor performs a copy protection method. The copy protection method comprises executing a software loop of a first software application in a first operating system. A first call is executed in the software loop to a code portion. A decrypted code portion of the first software application is executed in a second operating system in response to the first call. The code portion is decrypted in response to a successful validation of the first software application.

Abstract: A high-definition map system receives sensor data from vehicles traveling along routes and combines the data to generate a high definition map for use in driving vehicles, for example, for guiding autonomous vehicles. A pose graph is built from the collected data, each pose representing location and orientation of a vehicle. The pose graph is optimized to minimize constraints between poses. Points associated with surface are assigned a confidence measure determined using a measure of hardness/softness of the surface. A machine-learning-based result filter detects bad alignment results and prevents them from being entered in the subsequent global pose optimization. The alignment framework is parallelizable for execution using a parallel/distributed architecture. Alignment hot spots are detected for further verification and improvement. The system supports incremental updates, thereby allowing refinements of subgraphs for incrementally improving the high-definition map for keeping it up to date.