Abstract: A geometric approach may be used to detect objects on a road surface. A set of points within a region of interest between a first frame and a second frame are captured and tracked to determine a difference in location between the set of points in two frames. The first frame may be aligned with the second frame and the first pixel values of the first frame may be compared with the second pixel values of the second frame to generate a disparity image including third pixels. One or more subsets of the third pixels that have a value above a first threshold may be combined, and the third pixels may be scored and associated with disparity values for each pixel of the one or more subsets of the third pixels. A bounding shape may be generated based on the scoring.

Abstract: Systems and methods for dynamically generating a predicted similarity score for a pair of input sequences. A predicted similarity score for a pair of input sequences is determined based at least in part on at least one of a token-level similarity probability score for the pair of input sequences, a target region match indication for the pair of input sequences, a fuzzy match score for the pair of input sequences, a character-level match score for the pair of input sequences, one or more similarity ratio occurrence indicators for the pair of input sequences, and a harmonic mean score of the fuzzy match score for the pair of input sequences and the token-level similarity probability score for the pair of input sequences.

Abstract: A method for providing data for creating a digital map. The method includes: detecting surroundings sensor data of the surroundings during a measuring run of a physical system, preferably a vehicle, the surroundings sensor data capturing the surroundings in an at least partially overlapping manner, first surroundings sensor data including three-dimensional information, and second surroundings sensor data including two-dimensional information; extracting, with the aid of a first neural network situated in the physical system, at least one defined object from the first and second surroundings sensor data into first extracted data; and extracting, with the aid of a second neural network situated in the physical system, characteristic features including descriptors from the first extracted data into second extracted data, the descriptors being provided for a defined alignment of the second extracted data in a map creation process.

Abstract: An image detection apparatus includes: a display outputting an image; a memory storing one or more instructions; and a processor configured to execute the one or more instructions stored in the memory to: detect, by using a neural network, an additional information area in a first image output on the display; obtain style information of the additional information area from the additional information area; and detect, in a second image output on the display, an additional information area having style information different from the style information by using a model that has learned an additional information area having new style information generated based on the style information.

Abstract: A system uses a neural network based model to perform scene text recognition. The system achieves high accuracy of prediction of text from scenes based on a neural network architecture that uses double attention mechanism. The neural network based model includes a convolutional neural network component that outputs a set of visual features and an attention extractor neural network component that determines attention scores based on the visual features. The visual features and the attention scores are combined to generate mixed features that are provided as input to a character recognizer component that determines a second attention score and recognizes the characters based on the second attention score. The system trains the neural network based model by adjusting the neural network parameters to minimize a multi-class gradient harmonizing mechanism (GHM) loss. The multi-class GHM loss varies based on a level of difficulty of the sample.

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: Techniques are provided herein for retrieving images that correspond to a target subject matter within a target context. Although useful in a number of applications, the techniques provided herein are particularly useful in contextual product association and visualization. A method is provided to apply product images to a neural network. The neural network is configured to classify the products in the images. The images are associated with a context representing the combination of classified products in the images. These techniques leverage both seller-provided images of products and user-generated content, which potentially includes hundreds or thousands of images of the same or similar products as the seller-provided images. A graphical user interface is configured to permit a user to select the context of interest in which to visualize the products.

Abstract: The technology described herein is directed to a cross-domain training framework that iteratively trains a domain adaptive refinement agent to refine low quality real-world image acquisition data, e.g., depth maps, when accompanied by corresponding conditional data from other modalities, such as the underlying images or video from which the image acquisition data is computed. The cross-domain training framework includes a shared cross-domain encoder and two conditional decoder branch networks, e.g., a synthetic conditional depth prediction branch network and a real conditional depth prediction branch network. The shared cross-domain encoder converts synthetic and real-world image acquisition data into synthetic and real compact feature representations, respectively.

Abstract: A system for analyzing images includes a processing device includes a receiving module configured to receive an image, and an analysis module configured to apply the received image to a machine learning network and classify one or more features in the received image, the machine learning network configured to propagate image data through a plurality of convolutional layers, each convolutional layer of the plurality of convolutional layers including a plurality of filter channels, the machine learning network including a bottleneck layer configured to recognize an image feature based on a shape of an image component, The system also includes an output module configured to output characterization data that includes a classification of the one or more features.

Abstract: In a method for identification of an Internet meme, a plurality of sources is monitored for digital visual content comprising a visual moment and a caption. It is determined whether instances of digital visual content include a same visual moment. Provided the instances of digital visual content include the same visual moment, the instances of digital visual content including the same visual moment are identified as similar digital visual content. Each instance of the similar digital visual content is tracked. Provided a total number of instances of the similar digital visual content exceeds an Internet meme threshold, the similar digital visual content is identified as an Internet meme, wherein the same visual moment is a root visual moment and each caption corresponds to a different iteration of the Internet meme.

Abstract: A method for detecting a face synthetic image, an electronic device and a storage medium are provided. The technical solution includes inputting a face image to be detected into a pre-trained convolution neural network to obtain a raw image feature of the face image; inputting the raw image feature into a first full connected layer and a second full connected layer respectively to obtain a first feature vector corresponding to a face key point of the face image and a second feature vector corresponding to the face image; merging the first feature vector and the second feature vector to obtain a merged feature vector; inputting the merged feature vector to a third full connected layer to obtain a detection result of the face image.

Abstract: A vehicle system includes a lidar system that obtains an initial point cloud and obtains a dual density point cloud by implementing a first neural network and based on the initial point cloud. The dual density point cloud results from reducing point density of the initial point cloud outside a region of interest (ROI). Processing the dual density point cloud results in a detection result that indicates any objects in a field of view (FOV) of the lidar system. A controller obtains the detection result from the lidar system and controls an operation of the vehicle based on the detection result.

Abstract: A system and method for field extraction including determining a key position of a key in an electronic file, isolating candidate key values based on a distance from the key position, selecting a key value from the candidate key values based on an output of a trained neural network, and extracting the key and the key value from the electronic file, regardless of a key-value structure.

Abstract: A method and apparatus with pose estimation, where the method may include obtaining, using a depth network, a respective depth image for each of a plurality of successive input images, obtaining, using a pose network, respective image pose transformation matrices between images, of the successive input images, at adjacent time points, obtaining, based on initial pose information and the respective image pose transformation matrices, image pose information for each of the adjacent times, estimating final pose information dependent on the obtained image pose information, accumulating the image pose transformation matrices, calculating a pose loss value based on a result of comparing image position information, obtained from a result of the accumulating, and sensor position information obtained from a sensor. The pose and depth networks may be updated based on the pose loss value and a composite loss value dependent on the image pose transformation matrices and the input images.

Abstract: A method of multi-directional scene text recognition based on multi-element attention mechanism include: performing normalization processing for a text row/column image output from an external text detection module by a feature extractor, extracting a feature for the normalized image by using a deep convolutional neural network to acquire an initial feature map, adding a 2-dimensional directional positional encoding P to the initial feature map in order to output a multi-channel feature map, converting the multi-channel feature map output from a feature extractor by an encoder into a hidden representation, and converting the hidden representation output from the encoder into a recognized text by a decoder and using the recognized text as the output result.

Abstract: A method for parallel processing of retinal images includes: optimizing an objective function with a chaotic supply-demand algorithm to enhance a real retinal image; synthesizing a virtual retinal image by a hybrid image generation method; establishing a parallel multi-layer decomposed interval type-2 intuitionistic fuzzy convolutional neural network model based on the virtual retinal image and the enhanced real retinal image; and integrating outputs from a plurality of parallel multi-layer decomposed interval type-2 intuitionistic fuzzy convolutional neural network models as a final classification result.

Abstract: A text recognition method includes: acquiring an image including text information, the text information including M characters, M being a positive integer greater than 1; performing text recognition on the image to acquire character information about the M characters; recognizing reading direction information about each character in accordance with the character information about the M characters, the reading direction information being used to indicate a next character corresponding to a current character in a semantic reading order; and ranking the M characters in accordance with the reading direction information about the M characters to acquire a text recognition result of the text information.

Abstract: A system is provided for training an inferential model based on selected training vectors. During operation, the system receives training data comprising observations for a set of time-series signals gathered from sensors in a monitored system during normal fault-free operation. Next, the system divides the observations into N subgroups comprising non-overlapping time windows of observations. The system then selects observations with a local minimum value and a local maximum value for all signals from each subgroup to be training vectors for the inferential model. Finally, the system trains the inferential model using the selected training vectors. Note that by selecting observations with local minimum and maximum values to be training vectors, the system maximizes an operational range for the training vectors, which reduces clipping in estimates subsequently produced by the inferential model and thereby reduces false alarms.

Abstract: According to an embodiment, an information processing device include: one or more processors. The processors input data based on input data including first input data belonging to a first domain and second input data belonging to a second domain different from the first domain, to a first model, and acquire first output data indicating an execution result of a first task with the first model. The processors input data based on the input data to a second model, and acquire second output data indicating an execution result of a second task with the second model. The processors convert the first output data into first conversion data expressed in a form of an execution result of the second task. The processors generate supervised data of the second model for the first input data, based on the first conversion data and the second output data.

Abstract: An apparatus includes a camera and a processing circuit. The camera may be configured to capture images of an environment around a vehicle. The processing circuit may be configured to (i) perform automated number-plate recognition using the images, (ii) store a history of detected license plates, and (iii) search the history of detected license plates in response to receiving a request from a communication device of a vehicle user for information matching the request.