In one example, an apparatus comprises: a first sensor layer, including an array of pixel cells configured to generate pixel data; and one or more semiconductor layers located beneath the first sensor layer with the one or more semiconductor layers being electrically connected to the first sensor layer via interconnects. The one or more semiconductor layers comprises on-chip compute circuits configured to receive the pixel data via the interconnects and process the pixel data, the on-chip compute circuits comprising: a machine learning (ML) model accelerator configured to implement a convolutional neural network (CNN) model to process the pixel data; a first memory to store coefficients of the CNN model and instruction codes; a second memory to store the pixel data of a frame; and a controller configured to execute the codes to control operations of the ML model accelerator, the first memory, and the second memory.

A vehicular vision system includes at least one color camera disposed at a vehicle and having an image sensor operable to capture image data. A first system on chip (SoC) includes an image signal processor that receives captured image data and converts the received image data to converted data that is in a format suitable for machine vision processing. A second system on chip (SoC) receives captured image data and communicates display data to a display disposed in the vehicle, with the display data being in a format suitable for display of video images at the display. At startup of the vehicle, video images derived from the display data are displayed by the display within a time period following startup of the vehicle and machine vision data processing of converted data does not commence until after the display time period has elapsed following startup of the vehicle.

Methods, apparatus, systems, and articles of manufacture are disclosed to improve data training of a machine learning model using a field-programmable gate array (FPGA). An example system includes one or more computation modules, each of the one or more computation modules associated with a corresponding user, the one or more computation modules training first neural networks using data associated with the corresponding users, and FPGA to obtain a first set of parameters from each of the one or more computation modules, the first set of parameters associated with the first neural networks, configure a second neural network based on the first set of parameters, execute the second neural network to generate a second set of parameters, and transmit the second set of parameters to the first neural networks to update the first neural networks.

An imaging system has a imaging array on a semiconductor chip which also includes circuit the elements NPU and SRAM to rapidly identify target objects in the imaging data and output their high level representations with low power consumption.

Systems, apparatuses and methods may provide for technology that processes an inference workload in a first subset of layers of a neural network that prevents or inhibits data dependent branch operations, conducts an exit determination as to whether an output of the first subset of layers satisfies one or more exit criteria, and selectively bypasses processing of the output in a second subset of layers of the neural network based on the exit determination. The technology may also speculatively initiate the processing of the output in the second subset of layers while the exit determination is pending. Additionally, when the inference workloads include a plurality of batches, the technology may mask one or more of the plurality of batches from processing in the second subset of layers.

A method for the computer-aided training of an artificial neural network (ANN) for recognizing structural features on objects, by means of which method identified structural features on objects are recognizable rapidly and reliable. That is achieved by virtue of the fact that a convolutional neural network (CNN) having a multiplicity of neurons is used for the training of an ANN for feature recognition on objects. Said network comprises a multiplicity of convolutional and/or pooling layers for the extraction of information from images of individual objects. In this case, the images of the objects are respectively scaled or scaled up and/or down from layer to layer. During the scaling of the images information about the structural features of the objects is maintained, specifically independently of the scaling of the images.

A gaze point calculation apparatus includes: a first cache register, a multiplexer, an arithmetic unit assembly, and a state machine. The first cache register is configured to receive and store first coordinates and a plurality of calibration parameters required when second required are obtained through calculation according to the first coordinates. The state machine is configured to control the multiplexer to select each time at least one value from the first cache register and transmit same to the arithmetic unit assembly. The arithmetic unit assembly is configured to perform a preset operation on the at least one value received each time until the second coordinates are obtained, and output the second coordinates under control of the state machine.

Methods and systems are described herein for hosting and arbitrating algorithms for the generation of structured frames of data from one or more sources of unstructured input frames. A plurality of frames may be received from a recording device and a plurality of object types to be recognized in the plurality of frames may be determined. A determination may be made of multiple machine learning models for recognizing the object types. The frames may be sequentially input into the machine learning models to obtain a plurality of sets of objects from the plurality of machine learning models and object indicators may be received from those machine learning models. A set of composite frames with the plurality of indicators corresponding to the plurality of objects may be generated, and an output stream may be generated including the set of composite frames to be played back in chronological order.

A vehicle control device includes: a signal processing IC unit that outputs image processing data; a recognition processing IC unit that performs recognition processing of the external environment of a vehicle to output external environment data obtained through the recognition processing; a judgment processing IC unit that performs judgment processing for cruise control of the vehicle; a power management unit capable of controlling an on or off state of a recognition function of the external environment of the vehicle in the recognition processing IC unit according to the conditions of the vehicle; and a bypass path for enabling data communications from the signal processing IC unit to the judgment processing IC unit without performing the recognition processing of the external environment of the vehicle by the recognition processing IC unit.

Systems, apparatuses and methods may provide for technology that processes an inference workload in a first subset of layers of a neural network that prevents or inhibits data dependent branch operations, conducts an exit determination as to whether an output of the first subset of layers satisfies one or more exit criteria, and selectively bypasses processing of the output in a second subset of layers of the neural network based on the exit determination. The technology may also speculatively initiate the processing of the output in the second subset of layers while the exit determination is pending. Additionally, when the inference workloads include a plurality of batches, the technology may mask one or more of the plurality of batches from processing in the second subset of layers.