Generating gaze corrected images using bidirectionally trained network

At least one computer readable storage device comprising instructions that, when executed, cause one or more processors to at least:

• access a target angle and an input image, the input image including an eye in a first position;
• train an artificial neural network of a computing system by:
• generating an output image based on a training target angle and a training input image;
• calculating first weights of a model to reduce correction loss between the output image and a training target image;
• generating a reconstructed image based on a training input angle, the output image, and shared weights;
• calculating second weights of the model to reduce a reconstruction loss between the reconstructed image and the training input image;
• and
• adjust the shared weights based on the first weights and the second weights;
• generate a vector field with the neural network;
• and
• generate a gaze-adjusted image based on the vector field, the gaze-adjusted image including the eye in a second position.

A method comprising:

• accessing a target angle and an input image including an eye in a first position;
• bidirectionally training a neural network of a computing system by:
• generating an output image based on the target angle and the input image;
• calculating first weights of a model to reduce correction loss between the output image and a target image;
• generating a reconstructed image based on an input angle, the output image, and shared weights;
• calculating second weights of the model to reduce a reconstruction loss between the reconstructed image and the input image;
• and
• adjusting the shared weights based on the first weights and the second weights;
• generating a vector field with the neural network;
• and
• generating a gaze-adjusted image based on the vector field, the gaze-adjusted image including the eye in a second position.