Information processing apparatus, information processing method, vehicle, information processing server, and storage medium

A vehicle including an information processing apparatus,

• the information processing apparatus including:
• one or more processors;
• and
• a memory storing instructions which, when the instructions are executed by the one or more processors, cause the information processing apparatus to function as:
• an obtainment unit configured to obtain an actual image shot by an image capturing device;
• a recognition unit configured to recognize a target within the actual image by executing processing of a neural network that includes a hidden layer;
• and
• a processing unit configured to cause the neural network to perform learning by changing weighting coefficients for the neural network so that evaluation values based on a first evaluation function and a second evaluation function decrease,
• wherein the processing unit obtains intermediate outputs which correspond to the actual image and a computer graphics (CG) image and which are from the hidden layer of the neural network when each of the actual image and the CG image has been separately input to the neural network, the CG image showing a scene similar to the actual image, and
• the processing unit causes the neural network to perform learning with use of the evaluation values based on the first evaluation function and the second evaluation function, the first evaluation function causing the evaluation value to decrease as a difference between a recognition result from the recognition unit and training data decreases, the second evaluation function causing the evaluation value to decrease as a difference between the intermediate outputs corresponding to the actual image and the CG image decreases.

A vehicle comprising:

• an image capturing device;
• one or more processors;
• and
• a memory storing instructions which, when the instructions are executed by the one or more processors, cause the vehicle to function as:
• a recognition unit configured to recognize a target within an actual image shot by the image capturing device by executing processing of a neural network including a hidden layer,
• wherein the neural network is provided by an information processing server that includes a processing device configured to cause the neural network to perform learning by changing weighting coefficients for the neural network so that evaluation values based on a first evaluation function and a second evaluation function decrease,
• the information processing server causes the processing device to obtain intermediate outputs which correspond to the actual image and a computer graphics (CG) image and which are from the hidden layer of the neural network when each of the actual image and the CG image has been separately input to the neural network, the CG image showing a scene similar to the actual image, and
• the processing device causes the neural network to perform learning with use of the evaluation values based on the first evaluation function and the second evaluation function, the first evaluation function causing the evaluation value to decrease as a difference between a recognition result from the recognition unit and training data decreases, the second evaluation function causing the evaluation value to decrease as a difference between the intermediate outputs corresponding to the actual image and the CG image decreases.

An information processing server comprising:

• one or more processors;
• and
• a memory storing instructions which, when the instructions are executed by the one or more processors, cause the information processing server to function as:
• an obtainment unit configured to obtain, from an external information processing apparatus, an actual image shot by an image capturing device included in the external information processing apparatus;
• a recognition unit configured to recognize a target within the actual image by executing processing of a neural network that includes a hidden layer;
• and
• a processing unit configured to cause the neural network to perform learning by changing weighting coefficients for the neural network so that evaluation values based on a first evaluation function and a second evaluation function decrease,
• wherein the processing unit obtains intermediate outputs which correspond to the actual image and a computer graphics (CG) image and which are from the hidden layer of the neural network when each of the actual image and the CG image has been separately input to the neural network, the CG image showing a scene similar to the actual image, and
• the processing unit causes the neural network to perform learning with use of the evaluation values based on the first evaluation function and the second evaluation function, the first evaluation function causing the evaluation value to decrease as a difference between a recognition result from the recognition unit and training data decreases, the second evaluation function causing the evaluation value to decrease as a difference between the intermediate outputs corresponding to the actual image and the CG image decreases.

An information processing method that is executed by an information processing apparatus comprising:

• one or more processors;
• and a memory storing instructions which, when the instructions are executed by the one or more processors, cause the information processing apparatus to function as:
• an obtainment unit configured to obtain an actual image shot by an image capturing unit;
• a recognition unit configured to recognize a target within the actual image by executing processing of a neural network that includes a hidden layer;
• and a processing unit, the information processing method comprising:
• causing the neural network to perform learning by changing weighting coefficients for the neural network so that evaluation values based on a first evaluation function and a second evaluation function decrease,
• wherein, in the causing, intermediate outputs are obtained which correspond to the actual image and a computer graphics (CG) image and which are from the hidden layer of the neural network when each of the actual image and the CG image has been separately input to the neural network, the CG image showing a scene similar to the actual image, and
• in the causing, the neural network is caused to perform learning with use of the evaluation values based on the first evaluation function and the second evaluation function, the first evaluation function causing the evaluation value to decrease as a difference between a recognition result from the recognition unit and training data decreases, the second evaluation function causing the evaluation value to decrease as a difference between the intermediate outputs corresponding to the actual image and the CG image decreases.

An information processing method that is executed by an information processing server comprising:

• one or more processors;
• and a memory storing instructions which, when the instructions are executed by the one or more processors, cause the information processing server to function as:
• an obtainment unit configured to obtain, from an external information processing apparatus, an actual image shot by an image capturing device included in the external information processing apparatus;
• a recognition unit configured to recognize a target within the actual image by executing processing of a neural network that includes a hidden layer;
• and a processing unit, the information processing method comprising:
• causing the neural network to perform learning by changing weighting coefficients for the neural network so that evaluation values based on a first evaluation function and a second evaluation function decrease,
• wherein, in the causing, intermediate outputs are obtained which correspond to the actual image and a computer graphics (CG) image and which are from the hidden layer of the neural network when each of the actual image and the CG image has been separately input to the neural network, the CG image showing a scene similar to the actual image, and
• in the causing, the neural network is caused to perform learning with use of the evaluation values based on the first evaluation function and the second evaluation function, the first evaluation function causing the evaluation value to decrease as a difference between a recognition result from the recognition unit and training data decreases, the second evaluation function causing the evaluation value to decrease as a difference between the intermediate outputs corresponding to the actual image and the CG image decreases.

A non-transitory computer-readable storage medium storing instructions for causing a computer to function as each unit of an image processing apparatus,

• the image processing apparatus including:
• one or more processors;
• and a memory storing instructions which, when the instructions are executed by the one or more processors, cause the information processing apparatus to function as:
• an obtainment unit configured to obtain an actual image shot by an image capturing device;
• a recognition unit configured to recognize a target within the actual image by executing processing of a neural network that includes a hidden layer;
• and
• a processing unit configured to cause the neural network to perform learning by changing weighting coefficients for the neural network so that evaluation values based on a first evaluation function and a second evaluation function decrease,
• wherein the processing unit obtains intermediate outputs which correspond to the actual image and a computer graphics (CG) image and which are from the hidden layer of the neural network when each of the actual image and the CG image has been separately input to the neural network, the CG image showing a scene similar to the actual image, and
• the processing unit causes the neural network to perform learning with use of the evaluation values based on the first evaluation function and the second evaluation function, the first evaluation function causing the evaluation value to decrease as a difference between a recognition result from the recognition unit and training data decreases, the second evaluation function causing the evaluation value to decrease as a difference between the intermediate outputs corresponding to the actual image and the CG image decreases.