Abstract: Systems and methods are disclosed that relate to freespace detection using machine learning models. First data that may include object labels may be obtained from a first sensor and freespace may be identified using the first data and the object labels. The first data may be annotated to include freespace labels that correspond to freespace within an operational environment. Freespace annotated data may be generated by combining the one or more freespace labels with second data obtained from a second sensor, with the freespace annotated data corresponding to a viewable area in the operational environment. The viewable area may be determined by tracing one or more rays from the second sensor within the field of view of the second sensor relative to the first data. The freespace annotated data may be input into a machine learning model to train the machine learning model to detect freespace using the second data.

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: In various examples, data collection vehicles or machines may be equipped with one or more LiDAR sensors (and/or other sensors), and the LiDAR sensor(s) may be used to collect frames of LiDAR data representing various real-world conditions. The LiDAR data may be processed using one or more deep neural networks (DNNs) such as a transformer neural network to generate auto-labels representing detected dynamic obstacles of any designated class. Tracking may be applied to generate object tracks (tracklines), estimate velocity, and/or handle occlusions. In some embodiments, the object tracks may be refined based on geometry and/or confidence to improve their accuracy. In some embodiments, the auto-labels are classified to generate an estimated representation of quality, and auto-labels with at least a threshold quality score may be skipped during human labeling. As such, auto-label quality scores may be used to accelerate human validation of auto-labeled scenes by skipping high quality auto-labels.

Abstract: Apparatuses, systems, and techniques to generate labels of objects depicted with in an image. In at least one embodiment, one or more object labels are generated in association with an image frame based, at least in part, on one or more prior frames and one or more subsequent frames.

Abstract: An annotation pipeline may be used to produce 2D and/or 3D ground truth data for deep neural networks, such as autonomous or semi-autonomous vehicle perception networks. Initially, sensor data may be captured with different types of sensors and synchronized to align frames of sensor data that represent a similar world state. The aligned frames may be sampled and packaged into a sequence of annotation scenes to be annotated. An annotation project may be decomposed into modular tasks and encoded into a labeling tool, which assigns tasks to labelers and arranges the order of inputs using a wizard that steps through the tasks. During the tasks, each type of sensor data in an annotation scene may be simultaneously presented, and information may be projected across sensor modalities to provide useful contextual information. After all annotation tasks have been completed, the resulting ground truth data may be exported in any suitable format.

Abstract: An annotation pipeline may be used to produce 2D and/or 3D ground truth data for deep neural networks, such as autonomous or semi-autonomous vehicle perception networks. Initially, sensor data may be captured with different types of sensors and synchronized to align frames of sensor data that represent a similar world state. The aligned frames may be sampled and packaged into a sequence of annotation scenes to be annotated. An annotation project may be decomposed into modular tasks and encoded into a labeling tool, which assigns tasks to labelers and arranges the order of inputs using a wizard that steps through the tasks. During the tasks, each type of sensor data in an annotation scene may be simultaneously presented, and information may be projected across sensor modalities to provide useful contextual information. After all annotation tasks have been completed, the resulting ground truth data may be exported in any suitable format.

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 deep neural network(s) (DNN) may be used to perform panoptic segmentation by performing pixel-level class and instance segmentation of a scene using a single pass of the DNN. Generally, one or more images and/or other sensor data may be stitched together, stacked, and/or combined, and fed into a DNN that includes a common trunk and several heads that predict different outputs. The DNN may include a class confidence head that predicts a confidence map representing pixels that belong to particular classes, an instance regression head that predicts object instance data for detected objects, an instance clustering head that predicts a confidence map of pixels that belong to particular instances, and/or a depth head that predicts range values. These outputs may be decoded to identify bounding shapes, class labels, instance labels, and/or range values for detected objects, and used to enable safe path planning and control of an autonomous vehicle.

Abstract: An annotation pipeline may be used to produce 2D and/or 3D ground truth data for deep neural networks, such as autonomous or semi-autonomous vehicle perception networks. Initially, sensor data may be captured with different types of sensors and synchronized to align frames of sensor data that represent a similar world state. The aligned frames may be sampled and packaged into a sequence of annotation scenes to be annotated. An annotation project may be decomposed into modular tasks and encoded into a labeling tool, which assigns tasks to labelers and arranges the order of inputs using a wizard that steps through the tasks. During the tasks, each type of sensor data in an annotation scene may be simultaneously presented, and information may be projected across sensor modalities to provide useful contextual information. After all annotation tasks have been completed, the resulting ground truth data may be exported in any suitable format.

Abstract: In various examples, a deep neural network(s) (e.g., a convolutional neural network) may be trained to detect moving and stationary obstacles from RADAR data of a three-dimensional (3D) space, in both highway and urban scenarios. RADAR detections may be accumulated, ego-motion-compensated, orthographically projected, and fed into a neural network(s). The neural network(s) may include a common trunk with a feature extractor and several heads that predict different outputs such as a class confidence head that predicts a confidence map and an instance regression head that predicts object instance data for detected objects. The outputs may be decoded, filtered, and/or clustered to form bounding shapes identifying the location, size, and/or orientation of detected object instances. The detected object instances may be provided to an autonomous vehicle drive stack to enable safe planning and control of the autonomous vehicle.

Abstract: In various examples, automatic labeling of sensor representations for machine learning systems and applications. Systems and methods described herein may receive inputs for labeling one or more sensor representations of sensor data that represent objects or features, and then use those labels to automatically label additional sensor representations that also represent the same objects or features. For instance, and for an object, a user interface may include at least a map indicating a trajectory of the object, one or more sensor representations which represent the object, and a timeline indicating a time period for which the object was detected. A user may then use the user interface to label the object, such as with a bounding shape indicating a location of the object and/or with one or more attributes describing the object.

Abstract: A deep neural network(s) (DNN) may be used to perform panoptic segmentation by performing pixel-level class and instance segmentation of a scene using a single pass of the DNN. Generally, one or more images and/or other sensor data may be stitched together, stacked, and/or combined, and fed into a DNN that includes a common trunk and several heads that predict different outputs. The DNN may include a class confidence head that predicts a confidence map representing pixels that belong to particular classes, an instance regression head that predicts object instance data for detected objects, an instance clustering head that predicts a confidence map of pixels that belong to particular instances, and/or a depth head that predicts range values. These outputs may be decoded to identify bounding shapes, class labels, instance labels, and/or range values for detected objects, and used to enable safe path planning and control of an autonomous vehicle.

Abstract: In various examples, evaluating labeled training data for machine learning systems and applications is described herein. Systems and methods described herein may determine whether labels for training data are accurate based at least on additional labels for the training data that represent a consensus of how the training data should be labeled. For instance, sensor representations (e.g., images, point clouds, etc.) may initially be labeled using one or more automatic techniques (e.g., one or more machine learning models, one or more neural networks, one or more algorithms, etc.) and then verified and/or updated by users to generate first labels for the sensor representations. Additionally, copies of the sensor representations may also be labeled using additional users to generate second labels, where these second labels are then used to generate the consensus labels for the sensor representations. The consensus labels may then be used to evaluate the first labels.

Abstract: Approaches presented herein provide for the acceleration of a human review process, such as the review of annotations generated by a human labeler. Annotations (at least partially) generated by a human reviewer can be provided as input to a machine learning model trained to infer a probability of the annotations including at least one error. Annotations with a low probability of including an error can be approved automatically, while annotations with a high probability (e.g., above a threshold) of including an error can be directed for human review. In order to keep the human reviewer engaged, artificial errors may be introduced at various times based on various engagement criteria.

Abstract: In various examples, a deep neural network(s) (e.g., a convolutional neural network) may be trained to detect moving and stationary obstacles from RADAR data of a three dimensional (3D) space, in both highway and urban scenarios. RADAR detections may be accumulated, ego-motion-compensated, orthographically projected, and fed into a neural network(s). The neural network(s) may include a common trunk with a feature extractor and several heads that predict different outputs such as a class confidence head that predicts a confidence map and an instance regression head that predicts object instance data for detected objects. The outputs may be decoded, filtered, and/or clustered to form bounding shapes identifying the location, size, and/or orientation of detected object instances. The detected object instances may be provided to an autonomous vehicle drive stack to enable safe planning and control of the autonomous vehicle.

Abstract: In various examples, feature location identification for autonomous and semi-autonomous systems and applications is described herein. Systems and methods are disclosed that use LiDAR data to determine locations of road markings within an environment. For instance, the LiDAR data may be used to generate one or more images, such as a top-down image, birds-eye-view (BEV) image, and/or an intensity image, representing the environment. The image(s) may then be processed using one or more machine learning models that are configured to determine information associated with the road markings, such as bounding shapes (e.g., bounding boxes) indicating the locations of the road markings within the image(s), classifications associated with the road markings (e.g., the types of road markings), and/or any other information. The information may then be used to perform one or more processes, such as updating a map of the environment and/or navigating a vehicle within the environment.

Abstract: A deep neural network(s) (DNN) may be used to perform panoptic segmentation by performing pixel-level class and instance segmentation of a scene using a single pass of the DNN. Generally, one or more images and/or other sensor data may be stitched together, stacked, and/or combined, and fed into a DNN that includes a common trunk and several heads that predict different outputs. The DNN may include a class confidence head that predicts a confidence map representing pixels that belong to particular classes, an instance regression head that predicts object instance data for detected objects, an instance clustering head that predicts a confidence map of pixels that belong to particular instances, and/or a depth head that predicts range values. These outputs may be decoded to identify bounding shapes, class labels, instance labels, and/or range values for detected objects, and used to enable safe path planning and control of an autonomous vehicle.

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 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: In various examples, generating maps using first sensor data and then annotating second sensor data using the maps for autonomous systems and applications is described herein. Systems and methods are disclosed that automatically propagate annotations associated with the first sensor data generated using a first type of sensor, such as a LiDAR sensor, to the second sensor data generated using a second type of sensor, such as an image sensor(s). To propagate the annotations, the first type of sensor data may be used to generate a map, where the map represents the locations of static objects as well as the locations of dynamic objects at various instances in time. The map and annotations associated with the first sensor data may then be used to annotate the second sensor data and/or determine additional information associated with the objects represented by the second sensors data.