Abstract: In various examples, systems and methods are disclosed that preserve rich spatial information from an input resolution of a machine learning model to regress on lines in an input image. The machine learning model may be trained to predict, in deployment, distances for each pixel of the input image at an input resolution to a line pixel determined to correspond to a line in the input image. The machine learning model may further be trained to predict angles and label classes of the line. An embedding algorithm may be used to train the machine learning model to predict clusters of line pixels that each correspond to a respective line in the input image. In deployment, the predictions of the machine learning model may be used as an aid for understanding the surrounding environment—e.g., for updating a world model—in a variety of autonomous machine applications.

Abstract: A weighted average execution time associated with each execution stage of a plurality of execution stages used to process a plurality of frames in parallel is obtained. The processing of each of the plurality of frames is performed at each of the plurality of execution stages in a sequential order, starting with an initial execution stage and continuing with each subsequent execution stage. A first largest weighted average execution time associated with one of the plurality of execution stages is determined. A delay to the initial execution stage prior to processing a first next frame is applied. The delay is determined based on the first largest weighted average execution time.

Abstract: In various examples, to support training a deep neural network (DNN) to predict a dense representation of a 3D surface structure of interest, a training dataset is generated using a parametric mathematical modeling. A variety of synthetic 3D road surfaces may be generated by modeling a 3D road surface using varied parameters to simulate changes in road direction and lateral surface slope. In an example embodiment, a synthetic 3D road surface may be created by modeling a longitudinal 3D curve and expanding the longitudinal 3D curve to a 3D surface, and the resulting synthetic 3D surface may be sampled to form a synthetic ground truth projection image (e.g., a 2D height map). To generate corresponding input training data, a known pattern that represents which pixels may remain unobserved during 3D structure estimation may be generated and applied to a ground truth projection image to simulate a corresponding sparse projection image.

Abstract: In various embodiments, a finite aperture omni-directional camera is modeled by aligning a finite aperture lens and focal point with the omni-directional part of the projection. For example, each point on an image plane maps to a direction in camera space. For a spherical projection, the lens can be orientated along this direction and the focal point is picked along this direction at focal distance from the lens. For a cylindrical projection, the lens can be oriented along the projected direction on the two dimensional (2D) xz-plane, as the projection is not omni-directional in the y direction. The focal point is picked along the (unprojected) direction so its projection on the xz-plane is at focal distance from the lens. The final outgoing ray can be constructed by sampling of point on this oriented lens and shooting a ray from there through the focal point.

Abstract: Apparatuses, systems, and techniques to obtain metric data of a computing resource service provider. In at least one embodiment, metric data of one or more graphics processing unit (GPUs) is caused to be obtained from the one or more GPUs in an order from newest to oldest.

Abstract: Apparatuses, systems, and techniques to allocate portions of a storage to groups of processors. In at least one embodiment, an amount of storage to store data to be used by one or more computer programs, based at least in part, on an amount of processors to perform one or more portions of the one or more computer programs.

Abstract: Apparatuses, systems, and techniques for scheduling instructions in a cluster to guarantee GPU-CPU alignment for these instructions. In at least one embodiment, jobs are scheduled based on constraints on job sizes and job placement. In at least one embodiment, a processor comprises circuits to schedule instructions to be performed by processors based on latency of interconnects coupled to these processors.

Abstract: Apparatuses, systems, and techniques for selecting computing resources based on software programs scoring past performance of computing resources. In at least one embodiment, a processor comprising circuitry may cause software programs to be performed using computing resources based on software programs to score past performance of the one or more computing resources. In at least one embodiment, a processor selects a computing system to perform a software workload based on attributes of computer systems that are recorded over periods of time. In at least one embodiment, a processor comprises circuits to use attribute values of computer systems that are computed over periods of time to select computer systems to perform software workloads.

Abstract: A deep neural network(s) (DNN) may be used to detect objects from sensor data of a three dimensional (3D) environment. For example, a multi-view perception DNN may include multiple constituent DNNs or stages chained together that sequentially process different views of the 3D environment. An example DNN may include a first stage that performs class segmentation in a first view (e.g., perspective view) and a second stage that performs class segmentation and/or regresses instance geometry in a second view (e.g., top-down). The DNN outputs may be processed to generate 2D and/or 3D bounding boxes and class labels for detected objects in the 3D environment. As such, the techniques described herein may be used to detect and classify animate objects and/or parts of an environment, and these detections and classifications may be provided to an autonomous vehicle drive stack to enable safe planning and control of the autonomous vehicle.

Abstract: A method of decompression includes decompressing a compressed image according to a power curve to generate a partially decompressed image, wherein the compressed image is decompressed from a second bit depth that is lower than a first bit-depth at which the image was generated.

Abstract: In various examples, a two-dimensional (2D) and three-dimensional (3D) deep neural network (DNN) is implemented to fuse 2D and 3D object detection results for classifying objects. For example, regions of interest (ROIs) and/or bounding shapes corresponding thereto may be determined using one or more region proposal networks (RPNs)—such as an image-based RPN and/or a depth-based RPN. Each ROI may be extended into a frustum in 3D world-space, and a point cloud may be filtered to include only points from within the frustum. The remaining points may be voxelated to generate a volume in 3D world space, and the volume may be applied to a 3D DNN to generate one or more vectors. The one or more vectors, in addition to one or more additional vectors generated using a 2D DNN processing image data, may be applied to a classifier network to generate a classification for an object.

Abstract: A technique for block data transfer is disclosed that reduces data transfer and memory access overheads and significantly reduces multiprocessor activity and energy consumption. Threads executing on a multiprocessor needing data stored in global memory can request and store the needed data in on-chip shared memory, which can be accessed by the threads multiple times. The data can be loaded from global memory and stored in shared memory using an instruction which directs the data into the shared memory without storing the data in registers and/or cache memory of the multiprocessor during the data transfer.

Abstract: A computer implemented method of executing applications in a cloud server system is presented. The method comprises receiving a file identifier from a client device. The method also comprises receiving a file associated with the file identifier from a first server. Further, the method comprises accessing an application associated with the file from memory of the cloud server. Also, the method comprises executing by the cloud server the application using the file received from the first server. Finally, the method comprises streaming results from the executing the application as a video stream destined for the client device.

Abstract: One embodiment of the present invention sets forth a technique for performing spatial propagation. The technique includes generating a first directed acyclic graph (DAG) by connecting spatially adjacent points included in a set of unstructured points via directed edges along a first direction. The technique also includes applying a first set of neural network layers to one or more images associated with the set of unstructured points to generate (i) a set of features for the set of unstructured points and (ii) a set of pairwise affinities between the spatially adjacent points connected by the directed edges. The technique further includes generating a set of labels for the set of unstructured points by propagating the set of features across the first DAG based on the set of pairwise affinities.

Abstract: One embodiment of a computer-implemented method for processing ray tracing operations in parallel includes receiving a plurality of rays and a corresponding set of material shading instructions for each ray included in the plurality of rays for processing, wherein each ray represents a path from a light source to at least one point within a three-dimensional (3D) environment, and each corresponding set of material shading instructions is based at least in part on one or more material properties associated with at least one surface of at least one object included in the 3D environment; assigning each ray included in the plurality of rays to a different processing core included in a plurality of processing cores; and for each ray included in the plurality of rays, causing the processing core assigned to the ray to execute the corresponding set of material shading instructions on the ray to generate a color.

Abstract: In order to more accurately white balance an image, weightings can be determined for pixels of an image when computing an illuminant color value of the image and/or a scene. The weightings can be based at least in part on the Signal-to-Noise Ratio (SNR) of the pixels. The SNR may be actual SNR or SNR estimated from brightness levels of the pixels. SNR weighting (e.g., SNR adjustment) may reduce the effect of pixels with high noise on the computed illuminant color value. For example, one or more channel values of the illuminant color value can be determined based on the weightings and color values of the pixels. One or more color gain values can be determined based on the one or more channel values of the illuminant color value and used to white balance the image.

Abstract: LiDAR (light detection and ranging) and RADAR (radio detection and ranging) systems are commonly used to generate point cloud data for 3D space around vehicles, for such functions as localization, mapping, and tracking. Improved techniques for processing the point cloud data that has been collected are provided. The improved techniques include mapping one or more point cloud data points into a depth map, the one or more point cloud data points being generated using one or more sensors; determining one or more mapped point cloud data points within a bounded area of the depth map, and detecting, using one or more processing units and for an environment surrounding a machine corresponding to the one or more sensors, a location of one or more entities based on the one or more mapped point cloud data points.

Abstract: Systems and methods for cooling a datacenter are disclosed. In at least one embodiment, a hybrid coolant distribution unit (HCDU) provides first stage cooling of secondary coolant returning from at least one cold plate in a liquid to air heat exchanger (L2AHE) and provides second stage cooling in a liquid to liquid heat exchanger (L2LHE) for secondary coolant exiting an L2AHE using varying flow rates of a primary coolant based in part on a residual temperature that is determined for such secondary coolant exiting an L2AHE.

Abstract: In various examples, a deep neural network (DNN) may be used to detect and classify animate objects and/or parts of an environment. The DNN may be trained using camera-to-LiDAR cross injection to generate reliable ground truth data for LiDAR range images. For example, annotations generated in the image domain may be propagated to the LiDAR domain to increase the accuracy of the ground truth data in the LiDAR domain—e.g., without requiring manual annotation in the LiDAR domain. Once trained, the DNN may output instance segmentation masks, class segmentation masks, and/or bounding shape proposals corresponding to two-dimensional (2D) LiDAR range images, and the outputs may be fused together to project the outputs into three-dimensional (3D) LiDAR point clouds. This 2D and/or 3D information output by the DNN may be provided to an autonomous vehicle drive stack to enable safe planning and control of the autonomous vehicle.

Abstract: Systems and methods for cooling a datacenter are disclosed. In at least one embodiment, an absorption chiller includes a generator vessel to enable removal of heat from fluid returned from at least one computing component of the datacenter.