Abstract: A volumetric data structure models a particular volume representing the particular volume at a plurality of levels of detail. A first entry in the volumetric data structure includes a first set of bits representing voxels at a first level of detail, the first level of detail includes the lowest level of detail in the volumetric data structure, values of the first set of bits indicate whether a corresponding one of the voxels is at least partially occupied by respective geometry, where the volumetric data structure further includes a number of second entries representing voxels at a second level of detail higher than the first level of detail, the voxels at the second level of detail represent subvolumes of volumes represented by voxels at the first level of detail, and the number of second entries corresponds to a number of bits in the first set of bits with values indicating that a corresponding voxel volume is occupied.

Abstract: A view of geometry captured in image data generated by an imaging sensor is compared with a description of the geometry in a volumetric data structure. The volumetric data structure describes the volume at a plurality of levels of detail and includes entries describing voxels defining subvolumes of the volume at multiple levels of detail. The volumetric data structure includes a first entry to describe voxels at a lowest one of the levels of detail and further includes a number of second entries to describe voxels at a higher, second level of detail, the voxels at the second level of detail representing subvolumes of the voxels at the first level of detail. Each of these entries include bits to indicate whether a corresponding one of the voxels is at least partially occupied with the geometry. One or more of these entries are used in the comparison with the image data.

Abstract: Methods, systems, articles of manufacture and apparatus to generate digital scenes are disclosed. An example apparatus to generate labelled models includes a map builder to generate a three-dimensional (3D) model of an input image, a grouping classifier to identify a first zone of the 3D model corresponding to a first type of grouping classification, a human model builder to generate a quantity of placeholder human models corresponding to the first zone, a coordinate engine to assign the quantity of placeholder human models to respective coordinate locations of the first zone, the respective coordinate locations assigned based on the first type of grouping classification, a model characteristics modifier to assign characteristics associated with an aspect type to respective ones of the quantity of placeholder human models, and an annotation manager to associate the assigned characteristics as label data for respective ones of the quantity of placeholder human models.

Abstract: Methods, apparatus, systems and articles of manufacture to compress data are disclosed. An example apparatus includes a data slicer to split a dataset into a plurality of blocks of data; a data processor to select a first compression technique for a first block of the plurality of blocks of data based on first characteristics of the first block; and select a second compression technique for a second block of the plurality of blocks of data based on second characteristics of the second block; a first compressor to compress the first block using the first compression technique to generate a first compressed block of data; a second compressor to compress the second block using the second compression technique to generate a second compressed block of data; and a header generator to generate a first header identifying the first compression technique and a second header identifying the second compression technique.

Abstract: Systems and methods for distributed training of deep learning models are disclosed. An example local device to train deep learning models includes a reference generator to label input data received at the local device to generate training data, a trainer to train a local deep learning model and to transmit the local deep learning model to a server that is to receive a plurality of local deep learning models from a plurality of local devices, the server to determine a set of weights for a global deep learning model, and an updater to update the local deep learning model based on the set of weights received from the server.

Abstract: The present application discloses a computing device that can provide a low-power, highly capable computing platform for computational imaging. The computing device can include one or more processing units, for example one or more vector processors and one or more hardware accelerators, an intelligent memory fabric, a peripheral device, and a power management module. The computing device can communicate with external devices, such as one or more image sensors, an accelerometer, a gyroscope, or any other suitable sensor devices.

Abstract: Methods, apparatus, systems and articles of manufacture to store and access multi-dimensional data are disclosed. An example apparatus includes a memory; a memory allocator to allocate part of the memory for storage of a multi-dimensional data object; and a storage element organizer to: separate the multi-dimensional data into storage elements; store the storage elements in the memory, the stored storage elements being selectively executable; store starting memory address locations for the storage elements in an array in the memory, the array to facilitate selectable access of data of the stored elements; store a pointer for the array into the memory.

Abstract: Methods, apparatus, systems and articles of manufacture to perform dot product calculations using sparse vectors are disclosed. An example apparatus includes means for generating a mask vector based on a first logic operation on a difference vector and an inverse of a control vector, the control vector based on a first bitmap of a first sparse vector and a second bitmap of a second sparse vector; means for generating a first product of a third value from the first sparse vector and a fourth value from the second sparse vector, the third value based on (i) the mask vector and (ii) a second sparsity map based on the first sparse vector, the fourth value corresponding to (i) the mask vector and (ii) a second sparsity map corresponding to the second sparse vector; and means for adding the first product to a second product of a previous iteration.

Abstract: A disclosed example to operate a mobile camera includes recognizing a first feature in first sensor data in response to the first feature being detected in the first sensor data; transitioning the mobile camera from a first feature detection state to a second feature detection state in response to the recognizing of the first feature, the mobile camera to operate using higher power consumption in second feature detection state than in the first feature detection state; recognizing a second feature in second sensor data in the second feature detection state; and sending to an external device at least one of first metadata corresponding to the first feature or second metadata corresponding to the second feature.

Abstract: Methods, systems, apparatus, and articles of manufacture to reduce memory latency when fetching pixel kernels are disclosed. An example apparatus includes first interface circuitry to receive a first request from a hardware accelerator at a first time including first coordinates of a first pixel disposed in a first image block, second interface circuitry to receive a second request including second coordinates from the hardware accelerator at a second time after the first time, and kernel retriever circuitry to, in response to the second request, determine whether the first image block is in cache storage based on a mapping of the second coordinates to a block tag, and, in response to determining that the first image block is in the cache storage, access, in parallel, two or more memory devices associated with the cache storage to transfer a plurality of image blocks including the first image block to the hardware accelerator.

Abstract: The present application provides a method of corner detection and an image processing system for detecting corners in an image. The preferred implementation is in software using enabling and reusable hardware features in the underlying vector processor architecture. The advantage of this combined software and programmable processor datapath hardware is that the same hardware used for the FAST algorithm can also be readily applied to a variety of other computational tasks, not limited to image processing.

Abstract: Systems and methods are provided for image classification using histograms of oriented gradients (HoG) in conjunction with a trainer. The efficiency of the process is greatly increased by first establishing a bitmap which identifies a subset of the pixels in the HoG window as including relevant foreground information, and limiting the HoG calculation and comparison process to only the pixels included in the bitmap.

Abstract: Systems and methods for distributed training of deep learning models are disclosed. An example local device to train deep learning models includes a reference generator to label input data received at the local device to generate training data, a trainer to train a local deep learning model and to transmit the local deep learning model to a server that is to receive a plurality of local deep learning models from a plurality of local devices, the server to determine a set of weights for a global deep learning model, and an updater to update the local deep learning model based on the set of weights received from the server.

Abstract: A vector processor is disclosed including a variety of variable-length instructions. Computer-implemented methods are disclosed for efficiently carrying out a variety of operations in a time-conscious, memory-efficient, and power-efficient manner. Methods for more efficiently managing a buffer by controlling the threshold based on the length of delay line instructions are disclosed. Methods for disposing multi-type and multi-size operations in hardware are disclosed. Methods for condensing look-up tables are disclosed. Methods for in-line alteration of variables are disclosed.

Abstract: The present application relates generally to a parallel processing device. The parallel processing device can include a plurality of processing elements, a memory subsystem, and an interconnect system. The memory subsystem can include a plurality of memory slices, at least one of which is associated with one of the plurality of processing elements and comprises a plurality of random access memory (RAM) tiles, each tile having individual read and write ports. The interconnect system is configured to couple the plurality of processing elements and the memory subsystem. The interconnect system includes a local interconnect and a global interconnect.

Abstract: Methods, systems, apparatus and articles of manufacture to identify features within an image are disclosed herein. An example apparatus includes a horizontal cost (HCOST) engine to apply a first row of pixels of a macroblock to an input of a first HCOST unit, the first HCOST unit including a number of difference calculators; and a difference calculator engine to apply corresponding rows of pixels of a search window of a source image to corresponding ones of the number of difference calculators of the first HCOST unit, the corresponding ones of the number of difference calculators to calculate respective sums of absolute difference (SAD) values between (a) the first row of pixels of the macroblock and (b) the corresponding rows of pixels of the search window.

Abstract: Methods, apparatus, systems, and articles of manufacture are disclosed to improve convolution efficiency of a convolution neural network (CNN) accelerator. An example hardware accelerator includes a hardware data path element (DPE) in a DPE array, the hardware DPE including an accumulator, and a multiplier coupled to the accumulator, the multiplier to multiply first inputs including an activation value and a filter coefficient value to generate a first convolution output when the hardware DPE is in a convolution mode, and a controller coupled to the DPE array, the controller to adjust the hardware DPE from the convolution mode to a pooling mode by causing at least one of the multiplier or the accumulator to generate a second convolution output based on second inputs, the second inputs including an output location value of a pool area, at least one of the first inputs different from at least one of the second inputs.

Abstract: Example methods, apparatus, systems and articles of manufacture (e.g., physical storage media) to implement video surveillance with neural networks are disclosed. Example systems disclosed herein include a database to store records of operator-labeled video segments (e.g., as records of operator-labeled video segments). The operator-labeled video segments include reference video segments and corresponding reference event labels describing the video segments. Disclosed example systems also include a neural network including a first instance of an inference engine, and a training engine to train the first instance of the inference engine based on a training set of the operator-labeled video segments obtained from the database, the first instance of the inference engine to infer events from the operator-labeled video segments included in the training set.

Abstract: Methods, systems, articles of manufacture and apparatus to generate digital scenes are disclosed. An example apparatus to generate labelled models includes a map builder to generate a three-dimensional (3D) model of an input image, a grouping classifier to identify a first zone of the 3D model corresponding to a first type of grouping classification, a human model builder to generate a quantity of placeholder human models corresponding to the first zone, a coordinate engine to assign the quantity of placeholder human models to respective coordinate locations of the first zone, the respective coordinate locations assigned based on the first type of grouping classification, a model characteristics modifier to assign characteristics associated with an aspect type to respective ones of the quantity of placeholder human models, and an annotation manager to associate the assigned characteristics as label data for respective ones of the quantity of placeholder human models.

Abstract: Methods, apparatus, systems, and articles of manufacture to optimize pipeline execution are disclosed. An example apparatus includes a cost computation manager to determine a value associated with a first location of a first pixel of a first image and a second location of a second pixel of a second image by calculating a matching cost between the first location and the second location, and an aggregation generator to generate a disparity map including the value, and determine a minimum value based on the disparity map corresponding to a difference in horizontal coordinates between the first location and the second location.