A style transfer neural network may be used to generate stylized synthetic images, where real images provide the style (e.g., seasons, weather, lighting) for transfer to synthetic images. The stylized synthetic images may then be used to train a recognition neural network. In turn, the trained neural network may be used to predict semantic labels for the real images, providing recognition data for the real images. Finally, the real training dataset (real images and predicted recognition data) and the synthetic training dataset are used by the style transfer neural network to generate stylized synthetic images. The training of the neural network, prediction of recognition data for the real images, and stylizing of the synthetic images may be repeated for a number of iterations. The stylization operation more closely aligns a covariate of the synthetic images to the covariate of the real images, improving accuracy of the recognition neural network.

Various of the disclosed embodiments present systems and methods for distinguishing portions of a virtual model associated with a clothing article from portions of the virtual model not associated with the clothing article. Some embodiments facilitate quick and effective separation by employing a feature vector structure conducive to separation by a linear classifier. Such efficient separation may be especially beneficial in applications requiring the rapid scanning of large quantities of clothing while retaining high-fidelity representations of the clothing's geometry. Some embodiments further accommodate artist participation in the filtering process as well as scanning of articles from a variety of orientations and with a variety of supporting structures.

An image photographing method is provided, including: based on a photographing instruction, selecting at least one focus in a quadrangular image, and photographing at least one initial image based on the at least one focus, where the quadrangular image includes a first pair of opposite sides and a second pair of opposite sides, an included angle between the opposite sides of the first pair of opposite sides is less than that of the second pair of opposite sides, a first side is a shorter side of the first pair of opposite sides, the at least one focus includes a first focus, and the first focus is in the quadrangular image and in a partial region including the first side; adjusting a quadrangular image to a rectangular image; and obtaining a final image and outputting the final image.

An image processing apparatus including: a pattern detecting section which detects a specific pattern in an image; an image holding section which holds the image; an image conversion section which converts a pattern in the image into a pattern different from the specific pattern; and an output switching section which outputs an image having the pattern converted by the image conversion section for a region where the specific pattern is detected by the pattern detecting section, and outputs the image held by the image holding section for a region where the specific pattern is not detected.

Described is a system for object detection and recognition from a moving platform under adverse visual conditions. The system generates a predicted change regarding how a moving platform's surrounding change in an image captured from the moving platform. Based on an agreement of the predicted change and an actual input image obtained via an image sensor positioned on the moving platform and corrupted by a visual disturbance, a transformation-consistency map is generated. The transformation-consistency map and the actual input image are used to improve detection and recognition performance of a machine learning method in detecting at least one object in the moving platform's surroundings.

Various of the disclosed embodiments present systems and methods for distinguishing portions of a virtual model associated with a clothing article from portions of the virtual model not associated with the clothing article. Some embodiments facilitate quick and effective separation by employing a feature vector structure conducive to separation by a linear classifier. Such efficient separation may be especially beneficial in applications requiring the rapid scanning of large quantities of clothing while retaining high-fidelity representations of the clothing's geometry. Some embodiments further accommodate artist participation in the filtering process as well as scanning of articles from a variety of orientations and with a variety of supporting structures.

Disclosed is a distortion rectification method, comprising: taking a wide-angle photograph using a camera of a terminal; determining distortion regions and non-distortion regions in the wide-angle photograph; obtaining a target distortion region selected by a user; dividing the target distortion region into M grid regions of a first pre-set size, wherein M is an integer greater than or equal to one; and respectively performing distortion rectification on the M grid regions of the first pre-set size. Also disclosed is a terminal.

An image signal line driver circuit includes first to third source drivers and fourth to sixth source drivers, which are respectively cascade-connected. The output duration of data signals that are provide to these source drivers is increasingly short on the source drivers that are connected further downstream (that is, the amount of pixel data to be output to the next stage is increasingly small). This reduces the power consumption and heat generation of the overall device. Moreover, the phases of the data signals are shifted, thereby reducing EMI. In this way, when a plurality of image signal line driver circuits are cascade-connected, heat generation and power consumption in each driver circuit and/or EMI therebetween is reduced.

Techniques and systems are described for mapping 360-degree video data to a truncated square pyramid shape. A 360-degree video frame can include 360-degrees' worth of pixel data, and thus be spherical in shape. By mapping the spherical video data to the planes provided by a truncated square pyramid, the total size of the 360-degree video frame can be reduced. The planes of the truncated square pyramid can be oriented such that the base of the truncated square pyramid represents a front view and the top of the truncated square pyramid represents a back view. In this way, the front view can be captured at full resolution, the back view can be captured at reduced resolution, and the left, right, up, and bottom views can be captured at decreasing resolutions. Frame packing structures can also be defined for 360-degree video data that has been mapped to a truncated square pyramid shape.

A system and method for printing medical images is provided. The system includes a client device and a server in data communication with the client device. The client device includes a communication subsystem for receiving medical images from the server; a user interface device for displaying the medical images in an Internet browser and receiving a request to print at least one selected medical image; and a processor coupled to a memory for storing a plurality of instructions. The processor can be configured for generating a base model that includes at least one page required to print the at least one selected medical image, generating a printable model by, for each of the at least one selected medical image, rendering that selected image to an image element of a page of the base model; and transferring the printable model to a model printable by the Internet browser.