Provided are systems and methods that perform multi-task and/or multi-sensor fusion for three-dimensional object detection in furtherance of, for example, autonomous vehicle perception and control. In particular, according to one aspect of the present disclosure, example systems and methods described herein exploit simultaneous training of a machine-learned model ensemble relative to multiple related tasks to learn to perform more accurate multi-sensor 3D object detection. For example, the present disclosure provides an end-to-end learnable architecture with multiple machine-learned models that interoperate to reason about 2D and/or 3D object detection as well as one or more auxiliary tasks. According to another aspect of the present disclosure, example systems and methods described herein can perform multi-sensor fusion (e.g., fusing features derived from image data, light detection and ranging (LIDAR) data, and/or other sensor modalities) at both the point-wise and region of interest (ROI)-wise level, resulting in fully fused feature representations.

A method of determining a region of interest may include with a first image capturing device, determining a region of interest within a field of view common between the first image capturing device and a second image capturing device, the first image capturing device capturing an image at a lower resolution than the second image capturing device; determining a corner defining the region of interest; and upscaling the determined corner to match a corner within a field of view of an image captured by the second image capturing device.

In accordance with certain implementations of the present approach, a reduced, element-by-element, data set is transmitted between a robot having a sensor suite and a control system remote from the robot that is configured to display a representation of the environment local to the robot. Such a scheme may be useful in allowing a human operator remote from the robot to perform an inspection using the robot while the robot is on-site with an asset and the operator is off-site. In accordance with the present approach, an accurate representation of the environment in which the robot is situated is provided for the operator to interact with.

The disclosure relates to method and system for performing optimized image search. The method includes receiving an input image and user requirements with respect to an image search, identifying a sub-section from various sub-sections of the input image based on semantic parameters corresponding to each sub-section and the user requirements, and determining an optimal resolution of the sub-section based on a pixel density of various image formats derived for the sub-section. The method further includes identifying an optimal set of layers from a plurality of layers of an Artificial Neural Network (ANN) based image search model for performing the image search based on the semantic parameters for the sub-section, the optimal resolution of the sub-section, and historical data, and performing the image search to identify a set of output images similar to the sub-section based on a modified ANN based image search model comprising the optimal set of layers.

The disclosed technology is generally directed to carbon capture and storage. In one example of the technology, a first neural network is trained with synthetic data that is associated with seismic images of synthetic simulated subsurfaces. The first neural network extracts features from multiple resolutions of the seismic images of the synthetic simulated subsurfaces. The ground truth includes synthetic labels that indicate probabilities of potential carbon dioxide leakage pathways of the synthetic simulated subsurfaces. A seismic image of a first subsurface is received. At least the trained first neutral network is used to generate output labels that indicate probabilities of potential leakage pathways of carbon dioxide storage of the first subsurface.

A method and an apparatus are provided. The method includes receiving a video with a first plurality of frames having a first resolution; generating a plurality of warped frames from the first plurality of frames based on a first type of motion compensation; generating a second plurality of frames having a second resolution, wherein the second resolution is of higher resolution than the first resolution, wherein each of the second plurality of frames having the second resolution is derived from a subset of the plurality of warped frames using a convolutional network; and generating a third plurality of frames having the second resolution based on a second type of motion compensation, wherein each of the third plurality of frames having the second resolution is derived from a fusing a subset of the second plurality of frames.

The present application relates generally to identifying regions of interest in images, including but not limited to whole slide image region of interest identification, prioritization, de-duplication, and normalization via interpretable rules, nuclear region counting, point set registration, and histogram specification color normalization. This disclosure describes systems and methods for analyzing and extracting regions of interest from images, for example biomedical images depicting a tissue sample from biopsy or ectomy. Techniques directed to quality control estimation, granular classification, and coarse classification of regions of biomedical images are described herein. Using the described techniques, patches of images corresponding to regions of interest can be extracted and analyzed individually or in parallel to determine pixels correspond to features of interest and pixels that do not. Patches that do not include features of interest, or include disqualifying features, can be disqualified from further analysis. Relevant patches can analyzed and stored with various feature parameters.

This disclosure describes one or more implementations of a digital image semantic layout manipulation system that generates refined digital images resembling the style of one or more input images while following the structure of an edited semantic layout. For example, in various implementations, the digital image semantic layout manipulation system builds and utilizes a sparse attention warped image neural network to generate high-resolution warped images and a digital image layout neural network to enhance and refine the high-resolution warped digital image into a realistic and accurate refined digital image.

Concepts and technologies disclosed herein are directed to image classification attack mitigation. According to one aspect of the concepts and technologies disclosed herein, a system can obtain an original image and reduce a resolution of the original image to create a reduced resolution image. The system can classify the reduced resolution image and output a first classification. The system also can classify the original image via deep learning image classification and output a second classification. The system can compare the first classification and the second classification. In response to determining that the first classification and the second classification match, the system can output the second classification of the original image. In response to determining that the first classification and the second classification do not match, the system can output the first classification of the original image.

This disclosure describes one or more implementations of a digital image semantic layout manipulation system that generates refined digital images resembling the style of one or more input images while following the structure of an edited semantic layout. For example, in various implementations, the digital image semantic layout manipulation system builds and utilizes a sparse attention warped image neural network to generate high-resolution warped images and a digital image layout neural network to enhance and refine the high-resolution warped digital image into a realistic and accurate refined digital image.