A computer vision (CV) training system, includes: a supervised learning system to estimate a supervision output from one or more input images according to a target CV application, and to determine a supervised loss according to the supervision output and a ground-truth of the supervision output; an unsupervised learning system to determine an unsupervised loss according to the supervision output and the one or more input images; a weakly supervised learning system to determine a weakly supervised loss according to the supervision output and a weak label corresponding to the one or more input images; and a joint optimizer to concurrently optimize the supervised loss, the unsupervised loss, and the weakly supervised loss.

Discussed herein are devices, systems, and methods for organ mask generation. A device, system and method for organ mask generation including generating a synthetic centroid mask, identifying first and second intensity thresholds, in a first segmentation pass, setting (i) pixels of an image with intensities less than the first threshold to zero and (ii) pixels of the image corresponding to objects with centroids outside the synthetic centroid mask to zero, resulting an initial organ mask, in a second segmentation pass, setting pixels (i) with intensities less than the second threshold, the second threshold less than the first threshold to zero and (ii) setting pixels corresponding to objects with centroids outside the initial organ mask to zero, resulting in a second organ mask, and expanding and filling the second organ mask to generate an organ mask.

A first image can be acquired from a first sensor included in a vehicle and input to a deep neural network to determine a first bounding box for a first object. A second image can be acquired from the first sensor. Input latitudinal and longitudinal motion data from second sensors included in the vehicle corresponding to the time between inputting the first image and inputting the second image. A second bounding box can be determined by translating the first bounding box based on the latitudinal and longitudinal motion data. The second image can be cropped based on the second bounding box. The cropped second image can be input to the deep neural network to detect a second object. The first image, the first bounding box, the second image, and the second bounding box can be output.

Provided are systems and methods for realistic head turns and face animation synthesis. An example method includes receiving a source frame of a source video, where the source frame includes a head and a face of a source actor, generating source pose parameters corresponding to a pose of the head and a facial expression of the source actor; receiving a target image including a target head and a target face of a target person, determining target identity information associated with the target head and the target face of the target person, replacing source identity information in the source pose parameters with the target identity information to obtain further source pose parameters, and generating an output frame of an output video that includes a modified image of the target face and the target head adopting the pose of the head and the facial expression of the source actor.

Data is received that includes a feed of images of a plurality of objects passing in front of an inspection camera module forming part of a quality assurance inspection system. Thereafter, it is detected whether there is an object within each image. Based on this detection, images in which each object is detected that meet predefined object representation parameters are identified (on an object-by-object basis, etc.). The identified images are provided to a consuming application or process for quality assurance analysis. Related apparatus, systems, techniques and articles are also described.

Techniques are disclosed for systems and methods for water non-water segmentation of navigational imagery to assist in the autonomous navigation of mobile structures. An imagery based navigation system includes a logic device configured to communicate with an imaging module coupled to a mobile structure and/or configured to capture images of an environment about the mobile structure. The logic device may be configured to receive at least one image from the imaging module; determine a water/non-water segmented image based, at least in part, on the received at least one image, and generate a range chart corresponding to the environment about the mobile structure based, at least in part, on the determined water/non-water segmented image and/or the received at least one image.

Disclosed is a detection target positioning method and device. The method comprises: acquiring an original image and pre-processing the original image to obtain a gradation of each pixel in a target gradation image corresponding to a target region including a detection target; calculating first gradation sets corresponding to rows of pixels of the target gradation image and second gradation sets corresponding to columns of pixels of the target gradation image; and determining rows of two ends of the detection target in a column direction according to the first gradation sets, determining columns of two ends of the detection target in a row direction according to the second gradation sets, and determining a center of the detection target according to the row of two ends of the detection target in the column direction and the columns of two ends of the detection target in the row direction.

A method of image retargeting is provided. The method includes obtaining a source image, obtaining a target size for a retargeted image based on the source image, generating a two-dimensional importance map for the source image, generating, based on the two-dimensional importance map and the target size, a warping mesh having a distortion metric below a threshold value, determining whether a size of the warping mesh corresponds to the target size, and based on the size of the warping mesh being determined to correspond to the target size, rendering the retargeted image by applying the warping mesh to the source image.

The disclosure relates to systems, methods and programs for geometrically constrained, unsupervised training of convolutional autoencoders on unlabeled images for extracting eye landmarks. Disclosed systems for unsupervised deep learning of gaze estimation in eyes' image data are implementable in a computerized system. Disclosed methods include capturing an unlabeled image comprising the eye region of a user; and training a plurality of convolutional autoencoders on the unlabeled image comprising the eye region of a user using an initial geometrically regularized loss function to determine a plurality of eye landmarks.

Techniques are disclosed for providing a notification that a person is at a particular location. For example, a resident device may receive from a user device an image that shows a face of a first person, the image being captured by a first camera of the user device. The resident device may also receive, from another device having a second camera, a second image showing a portion of a face of a second person, the second camera having a viewable area showing a particular location. The resident device may determine a score indicating a level of similarity between a first set of characteristics associated with the face of the first person and a second set of characteristics associated with the face of a second person. The resident device may then provide to the user device a notification based on determining the score.