Visual stylization on stereoscopic images

A computer-implemented method, comprising:

• extracting a first feature map for a first source image and a second feature map for a second source image, the first and second source images corresponding to a first view and a second view of a stereoscopic image, respectively;
• determining, based on the first and second source images, a first unidirectional disparity from the first source image to the second source image;
• and
• generating a first target image and a second target image having a specified visual style by processing the first and second feature maps based on the first unidirectional disparity, wherein the first and second source images are comprised in a stereoscopic video which further comprises a third source image corresponding to the first view before the first source image and a fourth source image corresponding to the second view before the second source image, and wherein extracting the first and second feature maps comprises:
• extracting the first feature map based on the third source image and a third fused feature map corresponding to a third feature map for the third source image;
• and
• extracting the second feature map based on the fourth source image and a fourth fused feature map corresponding to a fourth feature map for the fourth source image.

A device, comprising:

• a processing unit;
• and
• a memory coupled to the processing unit and having instructions stored thereon which, when executed by the processing unit, cause the processing unit to perform acts comprising:
• extracting a first feature map for a first source image and a second feature map for a second source image;
• the first and second source images corresponding to a first view and a second view of a stereoscopic image, respectively;
• determining, based on the first and second source images, a first unidirectional disparity from the first source image to the second source image;
• and
• generating a first target image and a second target image having a specified visual style by processing the first and second feature maps based on the first unidirectional disparity, wherein generating the first and second target images comprises:
• warping, based on the first unidirectional disparity, the second feature map from the second view to the first view to obtain a first warped feature map;
• fusing the first feature map with the first warped feature map to obtain a first fused feature map;
• and
• generating the first and second target images based on the first fused feature map and the second feature map.

A computer-implemented method, comprising:

• updating a first sub-network of a learning network based on a reference stereoscopic image having a predetermined visual style and stereoscopic images each comprising a first source image corresponding to a first view and a second source image corresponding to a second view, to enable extrication of a first feature map for the first source image and a second feature map for the second source image using the first sub-network;
• updating a second sub-network of the learning network based on the stereoscopic images and a first unidirectional disparity from the first source image to the second source image, to enable generation of the respective first unidirectional disparity using the second sub-network based on the first and second source images;
• and
• updating a third sub-network of the learning network based on the stereoscopic images and the reference stereoscopic image, to enable generation of the reference stereoscopic image by processing the first and second feature maps using the third sub-network based on the first unidirectional disparity, wherein generation of the of the reference stereoscopic image comprises:
• warping, based on the first unidirectional disparity, the second feature map from the second view to the first view to obtain a first warped feature map;
• fusing the first feature map with the first warped feature map to obtain a first fused feature map;
• and
• generating the reference stereoscopic image based on the first fused feature map and the second feature map.

A device, comprising:

• a processing unit;
• and
• a memory coupled to the processing unit and having instructions stored thereon which;
• when executed by the processing unit, cause the processing unit to perform acts comprising:
• extracting a first feature map for a first source image and a second feature map for a second source image, the first and second source images corresponding to a first view and a second view of a stereoscopic image, respectively;
• determining, based on the first and second source images, a first unidirectional disparity from the first source image to the second source image;
• and
• generating a first target image and a second target image having a specified visual style by processing the first and second feature maps based on the first unidirectional disparity;
• wherein the first and second source images are comprised in a stereoscopic video which further comprises a third source image corresponding to the first view before the first source image and a fourth source image corresponding to the second view before the second source image, and wherein extracting the first and second feature maps comprises:
• extracting the first feature map based on the third source image and a third fused feature map corresponding to a third feature map for the third source image;
• and
• extracting the second feature map based on the fourth source image and a fourth fused feature map corresponding to a fourth feature map for the fourth source image.