Abstract: Systems, apparatuses, and methods for implementing a safety monitor framework for a safety-critical computer vision (CV) application are disclosed. A system includes a safety-critical CV application, a safety monitor, and a CV accelerator engine. The safety monitor receives an input image, test data, and a CV graph from the safety-critical CV application. The safety monitor generates a modified image by adding additional objects outside of the input image. The safety monitor provides the modified image and CV graph to the CV accelerator which processes the modified image and provides outputs to the safety monitor. The safety monitor determines the likelihood of erroneous processing of the original input image by comparing the outputs for the additional objects with a known good result. The safety monitor complements the overall fault coverage of the CV accelerator engine and covers faults only observable at the level of the CV graph.

Abstract: Systems, apparatuses, and methods for implementing a unified kernel virtual address space for heterogeneous computing are disclosed. A system includes at least a first subsystem running a first kernel, an input/output memory management unit (IOMMU), and a second subsystem running a second kernel. In order to share a memory buffer between the two subsystems, the first subsystem allocates a block of memory in part of the system memory controlled by the first subsystem. A first mapping is created from a first logical address of the kernel address space of the first subsystem to the block of memory. Then, the IOMMU creates a second mapping to map the physical address of that block of memory from a second logical address of the kernel address space of the second subsystem. These mappings allow the first and second subsystems to share buffer pointers which reference the block of memory.

Abstract: Systems, apparatuses, and methods for performing real-time video rendering with performance guaranteed power management are disclosed. A system includes at least a software driver, a power management unit, and a plurality of processing elements for performing rendering tasks. The system receives inputs which correspond to rendering tasks which need to be performed. The software driver monitors the inputs that are received and the number of rendering tasks to which they correspond. The software driver also monitors the amount of time remaining until the next video synchronization signal. The software driver determines which performance setting will minimize power consumption while still allowing enough time to finish the rendering tasks for the current frame before the next video synchronization signal. Then, the software driver causes the power management unit to provide this performance setting to the plurality of processing elements as they perform the rendering tasks for the current frame.

Abstract: An electronic device uses a tiling scheme selected from among a set of tiling schemes for processing instances of input data through a neural network. Each of the tiling schemes is associated with a different arrangement of portions into which instances of input data are divided for processing in the neural network. In operation, processing circuitry in the electronic device acquires information about a neural network and properties of the processing circuitry. The processing circuitry then selects a given tiling scheme from among a set of tiling schemes based on the information. The processing circuitry next processes instances of input data in the neural network using the given tiling scheme. Processing each instance of input data in the neural network includes dividing the instance of input data into portions based on the given tiling scheme, separately processing each of the portions in the neural network, and combining the respective outputs to generate an output for the instance of input data.

Abstract: Methods and devices are provided for processing image data on a sub-frame portion basis using layers of a convolutional neural network. The processing device comprises memory and a processor. The processor is configured to determine, for an input tile of an image, a receptive field via backward propagation and determine a size of the input tile based on the receptive field and an amount of local memory allocated to store data for the input tile. The processor determines whether the amount of local memory allocated to store the data of the input tile and padded data for the receptive field.

Abstract: A scheduler of an apparatus exposes an application programming interface (API) usable to specify quality-of-service (QoS) parameters, e.g., latency, throughput, and so forth. An application, for instance, specifies the QoS parameters for a workload to be processed using a hardware compute unit. The QoS parameters are employed by the scheduler as a basis to configure a partition within a hardware compute unit. The partition is configured such that processing resources that are available via the partition to process the workload comply with the specified quality-of-service.

Abstract: A system and method for automatically adjusting light conditions in a video conference environment are disclosed. A video conferencing system includes a camera to capture an image of a video conference participant and a computing device to evaluate and compensate for lighting conditions. A display device enables the participant to view other video conference participants. Video data captured by the camera is conveyed to the computing device. The computing device is configured to evaluate lighting conditions of a captured image and evaluate the lighting conditions for possible adjustment. Responsive to an evaluation of the lighting conditions, the computing device is configured to automatically generate a light border for display on the display device. The light border is composited with window display data received from a video conference application. The light border generated by the computing device is generated to create an amount of light that compensates for low light, or uneven light, conditions.

Abstract: Region-of-interest (ROI)-based image enhancement using a residual network, including: generating, based on an input image and a residual path of a residual network, a first output corresponding to a region-of-interest of the input image; generating, based on the input image and a skip path of the residual network, a second output; and generating an output image based on the first output and the second output.

Abstract: Methods and devices are provided for processing image data on a sub-frame portion basis using layers of a convolutional neural network. The processing device comprises memory and a processor. The processor is configured to receive frames of image data comprising sub-frame portions, schedule a first sub-frame portion of a first frame to be processed by a first layer of the convolutional neural network when the first sub-frame portion is available for processing, process the first sub-frame portion by the first layer and continue the processing of the first sub-frame portion by the first layer when it is determined that there is sufficient image data available for the first layer to continue processing of the first sub-frame portion. Processing on a sub-frame portion basis continues for subsequent layers such that processing by a layer can begin as soon as sufficient data is available for the layer.

Abstract: Systems, apparatuses, and methods for implementing a safety monitor framework for a safety-critical inference application are disclosed. A system includes a safety-critical inference application, a safety monitor, and an inference accelerator engine. The safety monitor receives an input image, test data, and a neural network specification from the safety-critical inference application. The safety monitor generates a modified image by adding additional objects outside of the input image. The safety monitor provides the modified image and neural network specification to the inference accelerator engine which processes the modified image and provides outputs to the safety monitor. The safety monitor determines the likelihood of erroneous processing of the original input image by comparing the outputs for the additional objects with a known good result. The safety monitor complements the overall fault coverage of the inference accelerator engine and covers faults only observable at the network level.

Abstract: A computer vision processor in an image cluster defines a fenced memory region (FMR) that controls access to image data stored in a first portion of a trusted memory region (TMR). The computer vision processor receives FMR requests from an application implemented in a processing cluster. The FMR requests are to access the image data in the first portion of the TMR. The computer vision processor selectively allows the requesting application to access the image data. In some cases, the computer vision processor acquires the image data and stores the image data in the first portion of the TMR, such as buffers in the TMR. A data fabric selectively permits the image processing application to access the data stored in the TMR based on whether the image cluster has opened or closed the FMR for the portion of the TMR.

Abstract: Systems, apparatuses, and methods for implementing a safety monitor framework for a safety-critical inference application are disclosed. A system includes a safety-critical inference application, a safety monitor, and an inference accelerator engine. The safety monitor receives an input image, test data, and a neural network specification from the safety-critical inference application. The safety monitor generates a modified image by adding additional objects outside of the input image. The safety monitor provides the modified image and neural network specification to the inference accelerator engine which processes the modified image and provides outputs to the safety monitor. The safety monitor determines the likelihood of erroneous processing of the original input image by comparing the outputs for the additional objects with a known good result. The safety monitor complements the overall fault coverage of the inference accelerator engine and covers faults only observable at the network level.

Abstract: A computer vision processor in an image cluster defines a fenced memory region (FMR) that controls access to image data stored in a first portion of a trusted memory region (TMR). The computer vision processor receives FMR requests from an application implemented in a processing cluster. The FMR requests are to access the image data in the first portion of the TMR. The computer vision processor selectively allows the requesting application to access the image data. In some cases, the computer vision processor acquires the image data and stores the image data in the first portion of the TMR, such as buffers in the TMR. A data fabric selectively permits the image processing application to access the data stored in the TMR based on whether the image cluster has opened or closed the FMR for the portion of the TMR.

Abstract: Methods and devices are provided for processing image data on a sub-frame portion basis using layers of a convolutional neural network. The processing device comprises memory and a processor. The processor is configured to receive frames of image data comprising sub-frame portions, schedule a first sub-frame portion of a first frame to be processed by a first layer of the convolutional neural network when the first sub-frame portion is available for processing, process the first sub-frame portion by the first layer and continue the processing of the first sub-frame portion by the first layer when it is determined that there is sufficient image data available for the first layer to continue processing of the first sub-frame portion. Processing on a sub-frame portion basis continues for subsequent layers such that processing by a layer can begin as soon as sufficient data is available for the layer.

Abstract: Systems, apparatuses, and methods for implementing a safety monitor framework for a safety-critical computer vision (CV) application are disclosed. A system includes a safety-critical CV application, a safety monitor, and a CV accelerator engine. The safety monitor receives an input image, test data, and a CV graph from the safety-critical CV application. The safety monitor generates a modified image by adding additional objects outside of the input image. The safety monitor provides the modified image and CV graph to the CV accelerator which processes the modified image and provides outputs to the safety monitor. The safety monitor determines the likelihood of erroneous processing of the original input image by comparing the outputs for the additional objects with a known good result. The safety monitor complements the overall fault coverage of the CV accelerator engine and covers faults only observable at the level of the CV graph.

Abstract: Systems, apparatuses, and methods for implementing a safety monitor framework for a safety-critical computer vision (CV) application are disclosed. A system includes a safety-critical CV application, a safety monitor, and a CV accelerator engine. The safety monitor receives an input image, test data, and a CV graph from the safety-critical CV application. The safety monitor generates a modified image by adding additional objects outside of the input image. The safety monitor provides the modified image and CV graph to the CV accelerator which processes the modified image and provides outputs to the safety monitor. The safety monitor determines the likelihood of erroneous processing of the original input image by comparing the outputs for the additional objects with a known good result. The safety monitor complements the overall fault coverage of the CV accelerator engine and covers faults only observable at the level of the CV graph.

Abstract: Systems, apparatuses, and methods for performing real-time video rendering with performance guaranteed power management are disclosed. A system includes at least a software driver, a power management unit, and a plurality of processing elements for performing rendering tasks. The system receives inputs which correspond to rendering tasks which need to be performed. The software driver monitors the inputs that are received and the number of rendering tasks to which they correspond. The software driver also monitors the amount of time remaining until the next video synchronization signal. The software driver determines which performance setting will minimize power consumption while still allowing enough time to finish the rendering tasks for the current frame before the next video synchronization signal. Then, the software driver causes the power management unit to provide this performance setting to the plurality of processing elements as they perform the rendering tasks for the current frame.

Abstract: Systems, apparatuses, and methods for performing real-time video rendering with performance guaranteed power management are disclosed. A system includes at least a software driver, a power management unit, and a plurality of processing elements for performing rendering tasks. The system receives inputs which correspond to rendering tasks which need to be performed. The software driver monitors the inputs that are received and the number of rendering tasks to which they correspond. The software driver also monitors the amount of time remaining until the next video synchronization signal. The software driver determines which performance setting will minimize power consumption while still allowing enough time to finish the rendering tasks for the current frame before the next video synchronization signal. Then, the software driver causes the power management unit to provide this performance setting to the plurality of processing elements as they perform the rendering tasks for the current frame.

Abstract: Systems, apparatuses, and methods for implementing a unified kernel virtual address space for heterogeneous computing are disclosed. A system includes at least a first subsystem running a first kernel, an input/output memory management unit (IOMMU), and a second subsystem running a second kernel. In order to share a memory buffer between the two subsystems, the first subsystem allocates a block of memory in part of the system memory controlled by the first subsystem. A first mapping is created from a first logical address of the kernel address space of the first subsystem to the block of memory. Then, the IOMMU creates a second mapping to map the physical address of that block of memory from a second logical address of the kernel address space of the second subsystem. These mappings allow the first and second subsystems to share buffer pointers which reference the block of memory.

Abstract: Systems, apparatuses, and methods for performing real-time video rendering with performance guaranteed power management are disclosed. A system includes at least a software driver, a power management unit, and a plurality of processing elements for performing rendering tasks. The system receives inputs which correspond to rendering tasks which need to be performed. The software driver monitors the inputs that are received and the number of rendering tasks to which they correspond. The software driver also monitors the amount of time remaining until the next video synchronization signal. The software driver determines which performance setting will minimize power consumption while still allowing enough time to finish the rendering tasks for the current frame before the next video synchronization signal. Then, the software driver causes the power management unit to provide this performance setting to the plurality of processing elements as they perform the rendering tasks for the current frame.