Abstract: An image processing method of the present disclosure may include receiving a scanned image, and processing the received image through an octave convolution-based neural network to output a high-quality image and an edge image for the received image. The octave convolution-based neural network may include a plurality of octave encoder blocks and a plurality of octave decoder blocks. Each octave encoder block may include an octave convolutional layer, and may be configured to output a high-frequency feature map and a low-frequency feature map for the image.

Abstract: A cardiac device comprises a memory arranged for receiving haemodynamic data, and a computer arranged for applying a cardiovascular model comprising a cardiac model and an arterial and venous blood circulation model using the data received in the memory, and for extracting therefrom at least one cardiac activity indicator (CI).

Abstract: Apparatuses and methods of manufacturing same, systems, and methods for generating a convolutional neural network (CNN) are described. In one aspect, a minimal CNN having, e.g., three or more layers is trained. Cascade training may be performed on the trained CNN to insert one or more intermediate layers until a training error is less than a threshold. When cascade training is complete, cascade network trimming of the CNN output from the cascade training may be performed to improve computational efficiency. To further reduce network parameters, convolutional filters may be replaced with dilated convolutional filters with the same receptive field, followed by additional training/fine-tuning.

Abstract: A method is performed by a computing system. The method includes receiving image data from an imaging device, and determining, using the image data, a plurality of image-space tools, each image-space tool associated with a tool of a plurality of tools, each tool controlled by a manipulator of a plurality of manipulators. The method further includes determining a first correspondence between a first image-space tool of the plurality of image-space tools and a first tool of the plurality of tools based on a first disambiguation setting associated with the first tool.

Abstract: The proposed model is a Variational Autoencoder having a learnable prior that is parametrized with a Tensor Train (VAE-TTLP). The VAE-TTLP can be used to generate new objects, such as molecules, that have specific properties and that can have specific biological activity (when a molecule). The VAE-TTLP can be trained in a way with the Tensor Train so that the provided data may omit one or more properties of the object, and still result in an object with a desired property.

Abstract: A medical image processing method performed by a computer, for measuring the spatial location of a point on the surface of a patient's body including: acquiring at least two two-dimensional image datasets, wherein each two-dimensional image dataset represents a two-dimensional image of at least a part of the surface which comprises the point, and wherein the two-dimensional images are taken from different and known viewing directions; determining the pixels in the two-dimensional image datasets which show the point on the surface of the body; and calculating the spatial location of the point from the locations of the determined pixels in the two-dimensional image datasets and the viewing directions of the two-dimensional images; wherein the two-dimensional images are thermographic images.

Abstract: A system and method for instance-level lane detection for autonomous vehicle control are disclosed.

Abstract: A method and a system automatically generate a digital representation of an annulus structure of a valve from a segmented digital representation of a human internal heart. The basis for the segmented digital representation is multi-slice computed tomography image data. The method includes automatically determining, for at least a first effective time point, based on a segmentation, i.e. labels, of a provided input segmented digital representation, a candidate plane, and/or a candidate orientation vector together with a candidate center point, arranged with respect to the input segmented digital representation for the first effective time point, and candidate points for the annulus structure are determined automatically. From the candidate points acting as support points, a candidate spline interpolation is generated which is then adapted based on the input segmented digital representation.

Abstract: Disclosed are methods and systems for: (i) sequentially illuminating a specimen with different spatial distributions of light, wherein each illumination causes an object embedded in the specimen to emit radiation in response to the light; (ii) for each different spatial distribution of illumination light, imaging the radiation emitted from the specimen from each of multiple sides of the specimen; and (iii) determining information about the object in the specimen based on the imaged radiation from each of the multiple sides for each of the different spatial distributions of illumination light.

Abstract: An embodiment of the present disclosure provides a weight precision configuration method, including: determining a pre-trained preset neural network including a plurality of layers each having a preset weight precision; reducing, based on a current threshold, the weight precision of at least one layer in the preset neural network to obtain a corrected neural network having a recognition rate greater than the current threshold; and reducing the weight precision of a layer includes: adjusting the weight precision of the layer; setting, if a termination condition is met, the weight precision of the layer to a corrected weight precision that is less than or equal to the preset weight precision of the layer; and returning, if the termination condition is not met, to the operation of adjusting the weight precision of the layer; and determining a final weight precision of each layer to obtain a final neural network.

Abstract: The present disclosure provides systems and methods that leverage neural networks for high resolution image segmentation. A computing system can include a processor, a machine-learned image segmentation model comprising a semantic segmentation neural network and an edge refinement neural network, and at least one tangible, non-transitory computer readable medium that stores instructions that cause the processor to perform operations. The operations can include obtaining an image, inputting the image into the semantic segmentation neural network, receiving, as an output of the semantic segmentation neural network, a semantic segmentation mask, inputting at least a portion of the image and at least a portion of the semantic segmentation mask into the edge refinement neural network, and receiving, as an output of the edge refinement neural network, the refined semantic segmentation mask.

Abstract: A system and method for facilitating DBS electrode trajectory planning using a machine learning (ML)-based feature identification scheme configured to identify and distinguish between various regions of interest (ROIs) and regions of avoidance (ROAs) in a patient's brain scan image. In one arrangement, standard orientation image slices as well as re-sliced images in non-standard orientations are provided in a labeled input dataset for training a CNN/ANN for distinguishing between ROIs and ROAs. Upon identification of the ROIs and ROAs in the patient's brain scan image, an optimal trajectory for implanting a DBS lead may be determined relative to a particular ROI while avoiding any ROAs.

Abstract: A software system for assisting a physician's diagnosis and reporting based on medical imaging includes software tools for pre-processing medical images, collecting findings, and automatically generating medical reports. A pre-processing software component generates an anatomical segmentation and/or computer-aided diagnosis based on an analysis of a medical image. A finding collecting software component displays the image, and facilitates rapid and efficient entry of associated findings by displaying a filtered list of templates associated with a selected region of the image and/or a computer-aided diagnosis. When the physician selects a template from the filtered list, the template may be displayed with entry options pre-filled based, e.g., on any computer-aided diagnosis. After the physician edits and/or confirms the entries, a report generation component uses the entries to generate a medical report.

Abstract: A method of and an Artificial Intelligence (AI) system for predicting hemodynamic parameters for a target vessel, in particular of an aorta, as well as to a computer-implemented method of training an AI unit of the AI system are disclosed. A vessel shape model of the target vessel and a corresponding flow profile of the target vessel are received. At least one hemodynamic parameter is predicted by the AI unit based on the received vessel shape model and the received flow profile. The AI unit is arranged and configured to predict at least one hemodynamic parameter based on a received vessel shape model and a received flow profile of the target vessel (aorta).

Abstract: Described herein are means for learning semantics-enriched representations via self-discovery, self-classification, and self-restoration in the context of medical imaging. Embodiments include the training of deep models to learn semantically enriched visual representation by self-discovery, self-classification, and self-restoration of the anatomy underneath medical images, resulting in a collection of semantics-enriched pre-trained models, called Semantic Genesis. Other related embodiments are disclosed.

Abstract: A machine receives a first set of global parameters from a global parameter server. The first set of global parameters includes data that weights one or more operands used in an algorithm that models an entity type. Multiple learner processors in the machine execute the algorithm using the first set of global parameters and a mini-batch of data known to describe the entity type. The machine generates a consolidated set of gradients that describes a direction for the first set of global parameters in order to improve an accuracy of the algorithm in modeling the entity type when using the first set of global parameters and the mini-batch of data. The machine transmits the consolidated set of gradients to the global parameter server. The machine then receives a second set of global parameters from the global parameter server, where the second set of global parameters is a modification of the first set of global parameters based on the consolidated set of gradients.

Abstract: Provided is a obtaining an excitation-emission matrix, wherein the excitation-emission matrix is measured with a spectrometer by: illuminating a biological tissue with stimulant light at a first wavelength to cause a first fluorescent emission of light by the biological tissue, measuring a first set of intensities of the first fluorescent emission of light at a plurality of different respective emission wavelengths, illuminating the biological tissue with stimulant light at a second wavelength to cause a second fluorescent emission of light by the biological tissue, and measuring a second set of intensities of the second fluorescent emission of light at a plurality of different respective emission wavelengths; and inferring a classification of the biological tissue or a concentration of a substance in the biological tissue with a multi-layer neural network or other machine learning model.

Abstract: Disclosed is a method for multi-center effect compensation based on a PET/CT intelligent diagnosis system. The method includes the following steps: estimating multi-center effect parameters of a test center B relative to a training center A by implementing a nonparametric mathematical method for data of the training center A and the test center B based on a location-scale model about additive and multiplicative multi-center effect parameters, and using the parameters to compensate the data of the test center B to eliminate a multi-center effect between the test center B and the training center A. According to the present disclosure, the multi-center effect between the training center A and the test center B can be compensated, so that the compensated data of the test center B can be used in the model trained by the training center A, and the generalization ability of the model is indirectly improved.

Abstract: Techniques are provided for deep neural network (DNN) identification of realistic synthetic images generated using a generative adversarial network (GAN). According to an embodiment, a system is described that can comprise a memory that stores computer executable components and a processor that executes the computer executable components stored in the memory. The computer executable components can comprise, a first extraction component that extracts a subset of synthetic images classified as non-real like as opposed to real-like, wherein the subset of synthetic images were generated using a GAN model. The computer executable components can further comprise a training component that employs the subset of synthetic images and real images to train a DNN network model to classify synthetic images generated using the GAN model as either real-like or non-real like.

Abstract: A predictor has a memory which stores at least one example for which an associated outcome is not known. The memory stores at least one decision tree comprising a plurality of nodes connected by edges, the nodes comprising a root node, internal nodes and leaf nodes. Individual ones of the nodes and individual ones of the edges each have an assigned module, comprising parameterized, differentiable operations, such that for each of the internal nodes the module computes a binary outcome for selecting a child node of the internal node. The predictor has a processor configured to compute the prediction by processing the example using a plurality of the differentiable operations selected according to a path through the tree from the root node to a leaf node.

Abstract: A system (100) includes a computer readable storage medium (122) with computer executable instructions (124), including: a biophysical simulator (126) configured to determine a fractional flow reserve value. The system further includes a processor (120) configured to execute the biophysical simulator (126), which employs machine learning to determine the fractional flow reserve value with spectral volumetric image data. The system further includes a display configured to display the determine fractional flow reserve value.

Abstract: Some embodiments provide a compiler for optimizing the implementation of a machine-trained network (e.g., a neural network) on an integrated circuit (IC). In some embodiments, the compiler determines whether sparsity requirements of channels implemented on individual cores are met on each core. If the sparsity requirement is not met, the compiler, in some embodiments, determines whether the channels of the filter can be rearranged to meet the sparsity requirements on each core and, based on the determination, either rearranges the filter channels or implements a solution to non-sparsity.

Abstract: A medical image data processing apparatus is provided and includes processing circuitry to receive medical image data in respect of at least one subject; receive non-image data; generate a filter based on the non-image data; and apply the filter to the medical image data, wherein the filter limits a region of the medical image data.

Abstract: Methods, apparatuses, and systems for providing real-time surgical assistance to a surgical robot using an imaging module, a context computer module, and a quantum analysis module are disclosed. The imaging module receives one or more data related to a patient from an intra-operative database and performs a first step analysis based on the received data. The context computer module determines a type of resource required for analysis based on data received from a historical database and the intra-operative database. The quantum analysis module receives data from the imaging module and the context computer module. The quantum analysis module receives real-time data from operation room (OR) equipment and performs quantum analysis on the received data and real-time data. The quantum analysis module facilitates real-time surgical assistance and recommendations to the surgical robot.

Abstract: The present invention relates to a system for efficient large-scale data distribution in a distributed and parallel processing environment. In particular, the present invention relates to global Top-k sparsification for low bandwidth networks. The present invention verifies that gTop-k S-SGD has nearly consistent convergence performance with S-SGD and evaluates the training efficiency of gTop-k on a cluster with 32 GPU machines which are inter-connected with 1 Gbps Ethernet. The experimental results show that the present invention achieves up to 2.7-12× higher scaling efficiency than S-SGD with dense gradients, and 1.1-1.7× improvement than the existing Top-k S-SGD.

Abstract: The disclosure relates to systems and methods for image processing. A trained deep learning model may be determined. An input image may be acquired. A processing result may be generated by processing the input image based on the trained deep learning model. The trained deep learning model may be obtained according to a process including: acquiring a preliminary deep learning model; acquiring a sample image; generating a plurality of sub-images based on the sample image; and training the preliminary deep learning model based on the plurality of sub-images to obtain the trained deep learning model.

Abstract: Various methods and systems are provided for transforming a style of an image into a different style while preserving content of the image. In one example, a method includes transforming a first image acquired via a medical imaging system into a second image based on visual characteristics of a third image using a system of deep neural networks configured to separate visual characteristics from content of an image, where the second image includes a content of the first image and the visual characteristics of the third image and the first and second images have different visual characteristics. The transformed second image may then be presented to a user.

Abstract: A method and a system for acquiring a 3D ultrasound image. The method includes receiving a request to capture a plurality of ultrasound image for a medical test corresponding to a medical condition. The method further includes determining a body part corresponding to the medical test. Further, the method includes identifying an imaging site particular to the medical test. Furthermore, the method includes providing a navigational guidance to the user in real time for positioning a handheld ultrasound device. Subsequently, the user is assisted to capture the plurality of ultrasound image of the imaging site in real time using deep learning. Further, the plurality of ultrasound images of the imaging site is captured. Finally, the method includes converting the plurality of ultrasound image to a 3-Dimensional (3D) ultrasound image in real time.

Abstract: One embodiment provides a method for training a machine-learning model to detect a location of a person's appendix, comprising: receiving, at the machine-learning model, a plurality of images, each image being a slice of a body taken by a CT scan; identifying, for each of the plurality of images, features of the appendix, wherein the identifying comprises analyzing a plurality of slices of each of the plurality of images and classifying each of the plurality of slices, into one of a plurality of classification groups, based upon a feature of the appendix within the slice; segmenting each of the plurality of image slices included in the one of the plurality of classification groups that classifies the slice as containing the appendix, thereby identifying probable locations of the appendix, via utilizing a probability mask for each of the probable locations.

Abstract: A medical scan triaging system is operable to generate a global abnormality probability for each of a plurality of medical scans by utilizing a computer vision model trained on a training set of medical scans. A triage probability threshold is determined based on user input to a client device. A first subset of the plurality of medical scans, designated for human review, is determined by identifying medical scans with a corresponding global abnormality probability that compares favorably to the triage probability threshold. A second subset of the plurality of medical scans, designated as normal, is determined by identifying ones of the plurality of medical scans with a corresponding global abnormality probability that compares unfavorably to the triage probability threshold. Transmission of the first subset of the plurality of medical scans to a plurality of client devices associated with a plurality of users is facilitated.

Abstract: A discriminator includes a common learning unit and a plurality of learning units that are connected to an output unit of the common learning unit. The discriminator is trained, using a plurality of data sets of a first image obtained by capturing an image of a subject that has developed a disease and an image data of a disease region in the first image, such that information indicating the disease region is output from a first learning unit in a case in which the first image is input to the common learning unit. In addition, the discriminator is trained, using a plurality of data sets of an image set obtained by registration between the first image and a second image whose type is different from the type of the first image, such that an estimated image of the second image is output from an output unit of a second learning unit.

Abstract: An embodiment of this application provides an image object detection method. The method may include obtaining a detection image, an n-level deep feature map framework, and an m-level non-deep feature map framework. The method may further include extracting deep feature from an (i?1)-level feature of the detection image using an i-level deep feature map framework, to obtain an i-level feature of the detection image. The method may further include extracting non-deep feature from a (j?1+n)-level feature of the detection image using a j-level non-deep feature map framework, to obtain a (j+n)-level feature of the detection image. The method may further include performing information regression operation on an a-level feature to an (m+n)-level feature of the detection image, to obtain an object type information and an object position information of an object in the detection image. The a is an integer less than n and greater than or equal to 2.

Abstract: Provided is a method based on deep neural network to extract appearance and geometry features for pulmonary textures classification, which belongs to the technical fields of medical image processing and computer vision. Taking 217 pulmonary computed tomography images as original data, several groups of datasets are generated through a preprocessing procedure. Each group includes a CT image patch, a corresponding image patch containing geometry information and a ground-truth label. A dual-branch residual network is constructed, including two branches separately takes CT image patches and corresponding image patches containing geometry information as input. Appearance and geometry information of pulmonary textures are learnt by the dual-branch residual network, and then they are fused to achieve high accuracy for pulmonary texture classification. Besides, the proposed network architecture is clear, easy to be constructed and implemented.

Abstract: Techniques are described for content-adaptive downsampling of digital images and videos for computer vision operations, such as semantic segmentation. A computer vision system comprises a memory, one or more processors operably coupled to the memory and a downsampling module configured for execution by the one or more processors to perform, based on a non-uniform sampling model trained to predict content-aware sampling parameters, downsampling input image data to generate downsampled image data. A segmentation module is configured for execution by the one or more processors to perform segmentation on the downsampled image to produce a segmentation result, such as a feature map that assigns pixels of the downsampled image data to object classes. An upsampling module is configured for execution by the one or more processors to perform upsampling according to the segmentation result to produce upsampled image data.

Abstract: An embodiment of the invention may include a method, computer program product and computer system for object detection and identification. The method, computer program product and computer system may include computing device which may receive an image from an imaging device. The image may be a medical image. The computing device may detect one or more potential indicators of disease in the image using a first algorithm and determine areas of potential disease in the image using an artificial intelligence algorithm. The computing device may determine a correlation between the determined areas of potential disease in the image and the one or more potential indicators of disease for the image. The computing device may, in response to determining a positive correlation, identify one or more of the potential indicators of disease for annotation and generate a report indicating one or more potential indicators of disease was found in the image.

Abstract: Disclosed are provided systems, methods, and apparatuses for implementing a multi-resolution neural network for use with imaging intensive applications including medical imaging. For example, a system having means to execute a neural network model formed from a plurality of layer blocks including an encoder layer block which precedes a plurality of decoder layer blocks includes: means for associating a resolution value with each of the plurality of layer blocks; means for processing via the encoder layer block a respective layer block input including a down-sampled layer block output processing, via decoder layer blocks, a respective layer block input including an up-sampled layer block output and a layer block output of a previous layer block associated with a prior resolution value of a layer block which precedes the respective decoder layer block; and generating the respective layer block output by summing or concatenating the processed layer block inputs.

Abstract: A joint position estimation device including a memory, and a processor connected to the memory. The processor executes a process including estimating, by a first DNN for which a first parameter determined by learning of the first DNN has been set, a body part region of the animal with respect to input image to be processed; and estimating, by the second DNN for which a second parameter determined by learning of the second DNN has been set, a first joint position and a second joint position in each of the body part region estimated by the first DNN and a plural body parts region in which a plurality of the body part regions are connected.

Abstract: Described herein are systems and methods related to image management. A system may include an instrument that includes an elongate body and an imaging device. The elongate body can be configured to be inserted into a luminal network. The imaging device may be positioned at a distal tip of the elongate body. The system may receive from the imaging device one or more images captured when the elongate body is within the luminal network. For each of the images, the system may determine one or more metrics that are indicative of a reliability of an image for localization of the distal tip of the elongate body within the luminal network. The system may determine a reliability threshold value for each of the one or more metrics. The system can utilize the one or more images based on whether the one or more metrics meet corresponding reliability threshold values.

Abstract: The present invention relates to a three-dimensional face image capturing method and comprises the steps of: capturing a face region of a user in a direction from a right chin to a left forehead; capturing the face region in a direction from a left forehead to a left chin; capturing the face region in a direction from a left chin to a right forehead; and capturing the face region in a direction from a right forehead to a center of face.

Abstract: An information processing apparatus and a method for volume data visualization is provided. The information processing apparatus stores an auto-encoder that includes an encoder network and a decoder network. The encoder network includes a loss function and a first plurality of neural network (NN) layers. The information processing apparatus inputs volume data to an initial NN layer of the first plurality of NN layers and generates a latent image as an output from a final NN layer of the first plurality of NN layers based on application of the encoder network on the input volume data. The information processing apparatus estimates a distance between the generated latent image and a reference image based on the loss function and updates the encoder network based on the estimated distance. Finally, the information processing apparatus outputs the updated encoder network as a trained encoder network based on the estimated distance being a minimum.

Abstract: The invention concerns a method for providing an evaluation means (60) for at least one optical application system (5) of a microscope-based application technology, wherein the following steps are performed, in particular each by an optical training system (4): performing an input detection (101) of at least one sample (2) according to the application technology in order to obtain at least one input record (110) of the sample (2) from the input detection (101), performing a target detection (102) of the sample (2) according to a training technology to obtain at least one target record (112) of the sample (2) from the target detection (102), the training technology being different from the application technology at least in that additional information (115) about the sample (2) is provided, training (130) of the evaluation means (60) at least on the basis of the input recording (110) and the target recording (112), in order to obtain a training information (200) of the evaluation means (60), in that vario

Abstract: An accession number correction system is operable to determine that an accession number of a received DICOM image does not link to any corresponding one of a plurality of medical reports. A query indicating medical report criteria, generated based on the first DICOM image, is transmitted to a report database, and a set of medical reports are received from the report database in response. One report of the set of medical reports that corresponds to the DICOM image is determined by performing a comparison function on the DICOM image and the one reports to generate a comparison value, and by determining the comparison value compares favorably to a comparison threshold. Updated report header data that includes the accession number of the first DICOM image is generated for the one report and is transmitted to the report database for storage.

Abstract: The purpose of the present disclosure is to provide a computer system, an animal diagnosis method, and a program in which the accuracy of animal diagnosis can be improved. The computer system acquires a visible light image of an animal imaged by a camera, compares the acquired visible light image with a normal visible light image of the animal and performs image analysis, identifies a species of the animal according to the result of the image analysis, identifies an abnormal portion of the animal according to the result of the image analysis, acquires environmental data of the animal, and diagnoses a condition of the animal according to the identified species, the identified abnormal portion and the acquired environmental data.

Abstract: Techniques are provided for deep neural network (DNN) identification of realistic synthetic images generated using a generative adversarial network (GAN). According to an embodiment, a system is described that can comprise a memory that stores computer executable components and a processor that executes the computer executable components stored in the memory. The computer executable components can comprise, a first extraction component that extracts a subset of synthetic images classified as non-real like as opposed to real-like, wherein the subset of synthetic images were generated using a GAN model. The computer executable components can further comprise a training component that employs the subset of synthetic images and real images to train a DNN network model to classify synthetic images generated using the GAN model as either real-like or non-real like.

Abstract: A method for determining optimized deep learning architecture includes receiving a plurality of training images and a plurality of real time images corresponding to a subject. The method further includes receiving, by a medical practitioner, a plurality of learning parameters comprising a plurality of filter classes and a plurality of architecture parameters. The method also includes determining a deep learning model based on the plurality of learning parameters and the plurality of training images, wherein the deep learning model comprises a plurality of reusable filters. The method further includes determining a health condition of the subject based on the plurality of real time images and the deep learning model. The method also includes providing the health condition of the subject to the medical practitioner.

Abstract: A machine trains a first neural network using a first set of images. Training the first neural network comprises computing a first set of weights for a first set of neurons. The machine, for each of one or more alpha values in order from smallest to largest, trains an additional neural network using an additional set of images. The additional set of images comprises a homographic transformation of the first set of images. The homographic transformation is computed based on the alpha value. Training the additional neural network comprises computing an additional set of weights for an additional set of neurons. The additional set of weights is initialized based on a previously computed set of weights. The machine generates a trained ensemble neural network comprising the first neural network and one or more additional neural networks corresponding to the one or more alpha values.

Abstract: A voxel feature learning network receives a raw point cloud and converts the point cloud into a sparse 4D tensor comprising three-dimensional coordinates (e.g. X, Y, and Z) for each voxel of a plurality of voxels and a fourth voxel feature dimension for each non-empty voxel. In some embodiments, convolutional mid layers further transform the 4D tensor into a high-dimensional volumetric representation of the point cloud. In some embodiments, a region proposal network identifies 3D bounding boxes of objects in the point cloud based on the high-dimensional volumetric representation. In some embodiments, the feature learning network and the region proposal network are trained end-to-end using training data comprising known ground truth bounding boxes, without requiring human intervention.

Abstract: Systems and methods are disclosed for processing an electronic image corresponding to a specimen. One method for processing the electronic image includes: receiving a target electronic image of a slide corresponding to a target specimen, the target specimen including a tissue sample from a patient, applying a machine learning system to the target electronic image to determine deficiencies associated with the target specimen, the machine learning system having been generated by processing a plurality of training images to predict stain deficiencies and/or predict a needed recut, the training images including images of human tissue and/or images that are algorithmically generated; and based on the deficiencies associated with the target specimen, determining to automatically order an additional slide to be prepared.

Abstract: A computer-implemented method for generating a composite image. The method includes iteratively optimizing an intermediate style transfer image using an initial style transfer image as a starting point based on a predefined loss function, original content features of a first input image, and original style features of a second input image; generating an optimized style transfer image after iteratively optimizing is performed for N times, N>1; and morphing the optimized style transfer image with the second input image to generate the composite image.

Abstract: An autonomous transport device system is enabled for one or more autonomous vehicles and autonomous surface delivery devices. Specific pickup and drop off zones for package delivery and user transport may be defined. The system leverages an artificial intelligence based learning algorithm to understand various environments. Packages may be dropped off in the geofenced areas. In some instances, packages may be stored in hidden areas that are purposely cached local to a likely delivery area. Some areas may be marked for pickup and drop off. Shippers may cache certain packages proximate to locations based on demand, joint distribution centers, and the presence of multiple transport devices including rovers, drones, UAVs, and autonomous vehicles.