Recognition device, method and storage medium

A method executed by a recognition device comprising storage configured to store:

• (i) a first recognition model created based on a first data set and a second recognition model created based on a second data set, the first data set comprising first recognition target data collected under a first condition and first correct data to be recognized from the first recognition target data, and the second data set comprising second recognition target data collected under a second condition and second correct data to be recognized from the second recognition target data, (ii) the first data set and the second data set, (iii) first and second tags indicative of the first condition, and (iv) third and fourth tags indicative of the second condition, the method comprising:
• acquiring a third data set comprising third recognition target data and third correct data to be recognized from the third recognition target data;
• executing recognition processing of the third recognition target data in the third data set by using each of the first recognition model and the second recognition model stored in the storage;
• computing a first recognition accuracy of the first recognition model by comparing a result of the recognition processing obtained by using the first recognition model with the third correct data;
• computing a second recognition accuracy of the second recognition model by comparing a result of the recognition processing obtained by using the second recognition model with the third correct data;
• using the computed first recognition accuracy as an accuracy of each of the first and the second tags;
• using the second computed second recognition accuracy as an accuracy of each of the third and the fourth tags;
• acquiring a representative value based on the first recognition accuracy and the second recognition accuracy when the first tag is the same as the third tag;
• extracting a particular tag of the first to the fourth tags based on the acquired representative value as an accuracy of each of the first and the third tags, the first recognition accuracy as the accuracy of the second tag, and the second recognition accuracy as the accuracy of the fourth tag;
• and
• creating a third recognition model based on the acquired third data set and a data set corresponding to the extracted particular tag among the first data set and the second data set.

A non-transitory computer-readable storage medium having stored thereon a computer program which is executable by a computer using storage configured to store:

• (i) a first recognition model created based on a first data set and a second recognition model created based on a second data set, the first data set comprising first recognition target data collected under a first condition and first correct data to be recognized from the first recognition target data, and the second data set comprising second recognition target data collected under a second condition and second correct data to be recognized from the second recognition target data, (ii) the first data set and the second data set, (iii) first and second tags indicative of the first condition, and (iv) third and fourth tags indicative of the second condition, the computer program comprising instructions capable of causing the computer to execute functions of:
• acquiring a third data set comprising third recognition target data and third correct data to be recognized from the third recognition target data;
• executing recognition processing of the third recognition target data in the third data set by using each of the first recognition model and the second recognition model stored in the storage;
• computing a first recognition accuracy of the first recognition model by comparing a result of the recognition processing obtained by using the first recognition model with the third correct data;
• computing a second recognition accuracy of the second recognition model by comparing a result of the recognition processing obtained by using the second recognition model with the third correct data;
• using the computed first recognition accuracy as an accuracy of each of the first and the second tags;
• using the second computed second recognition accuracy as an accuracy of each of the third and the fourth tags;
• acquiring a representative value based on the first recognition accuracy and the second recognition accuracy when the first tag is the same as the third tag;
• extracting a particular tag of the first to the fourth tags based on the acquired representative value as an accuracy of each of the first and the third tags, the first recognition accuracy as the accuracy of the second tag, and the second recognition accuracy as the accuracy of the fourth tag;
• and
• creating a third recognition model based on the acquired third data set and a data set corresponding to the extracted particular tag among the first data set and the second data set.