Image processing device, image processing method, and storage medium for adjusting image parameters

An image processing method comprising:

• acquiring an input image;
• dividing the input image into a plurality of partial images to obtain a partial image sequence;
• converting the partial image sequence into a token sequence by respectively converting each of the partial images includes in the partial image sequence into a token having a fixed-dimensional vector;
• adding a class token having a vector of the same dimension as the token to the token sequence;
• obtaining an encoded representation sequence based on an attention map indicating a degree of association between tokens and the token sequence;
• obtaining a feature vector from the encoded representation sequence obtained by the encoding;
• learning to adjust at least parameters for converting the partial image sequence into a token sequence unit and for obtaining the encoded representation sequence to reduce an attention loss value, wherein the attention loss value corresponds to an error between a target value of the degree of association between the partial images and a value indicated by the attention map calculated by the encoding;
• and
• acquiring the encoded representation sequence by updating the token sequence to which the class token is added based on the attention map.

A non-transitory computer-readable storage medium storing a computer program including instructions for executing following processes:

• acquiring an input image;
• dividing the input image into a plurality of partial images to obtain a partial image sequence;
• converting the partial image sequence into a token sequence by respectively converting each of the partial images includes in the partial image sequence into a token having a fixed-dimensional vector;
• adding a class token having a vector of the same dimension as the token to the token sequence;
• obtaining an encoded representation sequence based on an attention map indicating a degree of association between tokens and the token sequence;
• obtaining a feature vector from the encoded representation sequence obtained by the encoding;
• and
• learning to adjust at least parameters for converting the partial image sequence into a token sequence unit and for obtaining the encoded representation sequence to reduce an attention loss value, wherein the attention loss value corresponds to an error between a target value of the degree of association between the partial images and a value indicated by the attention map calculated by the encoding;
• and
• acquiring the encoded representation sequence by updating the token sequence to which the class token is added based on the attention map.

An image processing device comprising:

• at least one processor;
• and a memory coupled to the at least one processor, the memory storing instructions that, when executed by the at least one processor, cause the at least one processor to:
• acquire an input image;
• divide the input image into a plurality of partial images to obtain a partial image sequence;
• convert the partial image sequence into a token sequence by respectively converting each of the partial images includes in the partial image sequence into a token having a fixed-dimensional vector;
• obtain an encoded representation sequence based on an attention map indicating a degree of association between tokens and the token sequence;
• obtain a feature vector from the encoded representation sequence;
• adjust at least parameters for converting the partial image sequence into a token sequence unit and for obtaining the encoded representation sequence to reduce an attention loss value, wherein the attention loss value corresponds to an error between a target value of the degree of association between the partial images and a value indicated by the attention map;
• and
• synthesize a region unnecessary for face authentication with a face image and calculate the attention loss value for the face image that has undergone image synthesis.

An image processing device comprising:

• at least one processor;
• and a memory coupled to the at least one processor, the memory storing instructions that, when executed by the at least one processor, cause the at least one processor to:
• acquire an input image;
• divide the input image into a plurality of partial images to obtain a partial image sequence;
• convert the partial image sequence into a token sequence by respectively converting each of the partial images includes in the partial image sequence into a token having a fixed-dimensional vector;
• obtain an encoded representation sequence based on an attention map indicating a degree of association between tokens and the token sequence;
• obtain a feature vector from the encoded representation sequence;
• adjust at least parameters for converting the partial image sequence into a token sequence unit and for obtaining the encoded representation sequence to reduce an attention loss value, wherein the attention loss value corresponds to an error between a target value of the degree of association between the partial images and a value indicated by the attention map;
• estimate rectangle information and detection likelihood of a face in an image on the basis of the feature vector;
• calculate a detection loss value on the basis of the rectangle information and the detection likelihood, and
• adjust at least the parameters of the feature vector so that the detection loss value and the attention loss value are reduced.