Artificial neural networks (ANNs) are computing systems that imitate a human brain by learning to perform tasks by considering examples. These ANNs are typically created by connecting several layers of neural units using connections, where each neural unit is connected to every other neural unit either directly or indirectly to create fully connected layers within the ANN. However, by representing an artificial neural network utilizing paths from an input of the ANN to an output of the ANN, a complexity of the ANN may be reduced, and the ANN may be trained and implemented in a much faster manner when compared to fully connected layers within the ANN. More specifically, the ANN may be trained sparse from scratch in order to avoid a more expensive procedure of training the ANN and compressing it afterwards.

The present invention relates to a method and system for accurately predicting future trajectories of observed objects in dense and ever-changing city environments. More particularly, the present invention relates to substantially continuously tracking and estimating the future movements of an observed object. As an example, an observed object may be a moving vehicle, for example along a path or road. Aspects and/or embodiments seek to provide an end to end method and system for substantially continuously tracking and predicting future movements of a newly observed object, such as a vehicle, using motion prior data extracted from map data.

Presented herein are embodiments of energy-based models (EBMs), which may be trained via embodiments of a multistage coarse-to-fine expanding and sampling strategy. Embodiments of the training methodology start with learning a coarse-level EBM from images at low resolution and then gradually transits to learn a finer-level EBM from images at higher resolution by expanding the energy function as the learning progresses. Embodiments are computationally efficient with smooth learning and sampling. Tested embodiments achieved the best performance on image generation amongst all EBMs and successfully synthesized high-fidelity images. Embodiments may also be used for image restoration and out-of-distribution detection. Framework embodiments may be further generalized for one-sided unsupervised image-to-image translation and beat baseline methods in terms of model size and training budget. Also presented herein are embodiments of a gradient-based generative saliency methodology to interpret the translation dynamics.

A method and an apparatus for determining a vehicle speed computing a probability distribution of action intentions based on observation information of a surrounding object. Then, a probability redistribution of the different action intentions is computed based on travel times for the vehicle to travel from a current position to risk areas corresponding to the different action intentions, motion status variations of the surrounding object with the different action intentions are predicted based on the travel times for the vehicle to travel to the risk areas corresponding to the different action intentions. Finally, the travelling speed of the vehicle is determined based on the probability redistribution of the different action intentions, the motion status variations of the surrounding object with the different action intentions, and motion status variations of the vehicle under different travelling speed control actions.

Embodiments of this application disclose methods, systems, and devices for image region localization and medical image processing. In one aspect, a method comprises acquiring three-dimensional images of a target body part of a patient. The three-dimensional images comprise a plurality of magnetic resonant imaging (MRI) modalities. The method comprises registering a first image set of a first modality with a second image set of a second modality. After the registering, image features of the three-dimensional images are extracted. The image features are fused to obtain fused features. The method also comprises determining voxel types corresponding to voxels in the three-dimensional images according to the fused features. The method also comprises selecting, from the three-dimensional images, target voxels having a preset voxel type, obtaining position information of the target voxels, and localizing a target region within the target body part based on the position information of the target voxels.

A system including one or more processors and one or more non-transitory computer-readable media storing computing instructions configured to run on the one or more processors and perform receiving a title of an item associated with an online catalog; interpreting, using a natural language model, one or more attributes of the predetermined set of attributes; determining a first title quality score for the title based on a first rule; determining a second title quality score for the title based on a second rule; determining an aggregated title quality score for the title based on at least the first title quality score and the second title quality score; generating a content quality list for the title; and sending instructions to display, on a user interface of an electronic device, a content quality dashboard comprising the content quality list for the title of the item. Other embodiments are disclosed.

A method and an apparatus for determining a vehicle speed computing a probability distribution of action intentions based on observation information of a surrounding object. Then, a probability redistribution of the different action intentions is computed based on travel times for the vehicle to travel from a current position to risk areas corresponding to the different action intentions, motion status variations of the surrounding object with the different action intentions are predicted based on the travel times for the vehicle to travel to the risk areas corresponding to the different action intentions. Finally, the travelling speed of the vehicle is determined based on the probability redistribution of the different action intentions, the motion status variations of the surrounding object with the different action intentions, and motion status variations of the vehicle under different travelling speed control actions.

A method for identifying distracted pedestrians. The method includes determining operating conditions of a vehicle using a plurality of vehicle controllers. Pedestrian parameters for a pedestrian in a vicinity of the vehicle are acquired using a plurality of vehicle sensors. The pedestrian parameters include at least one of face positions, body positions, gait and hand gestures. Information related to an environment surrounding the vehicle is acquired. Pedestrian awareness level is determined based on the acquired pedestrian parameters and based on the information related to the environment surrounding the vehicle. A determination is made whether the pedestrian awareness level is below a predefined threshold. The pedestrian is classified as distracted, in response to determining that the pedestrian awareness level is below the predefined threshold.

Described is a system for detecting and correcting perception errors in a perception system. In operation, the system generates a list of detected objects from perception data of a scene, which allows for the generation of a list of background classes from backgrounds in the perception data associated with the list of detected objects. For each detected object in the list of detected objects, a closest background class is identified from the list of background classes. Vectors can then be used to determine a semantic feature, which is used to identify axioms. An optimal perception parameter is then generated, which is used to adjust perception parameters in the perception system to minimize perception errors.

Computer vision systems and methods for end-to end training of neural networks are provided. The system generates a fixed point algorithm for dual-decomposition of a maximum-a-posteriori inference problem and trains the convolutional neural network and a conditional random field with the fixed point algorithm and a plurality of images of a dataset to learn to perform semantic image segmentation. The system can segment an attribute of an image of the dataset by the trained neural network and the conditional random field.