Method and device for training image recognition model, and storage medium

A device for training an image recognition model, comprising:

• one or more processors;
• and
• a non-transitory computer-readable storage medium storing instructions executable by the one or more processors, wherein the one or more processors are configured to:
• train an image recognition model by using a tagged image sample set;
• recognize multiple to-be-tagged image samples in a to-be-tagged image sample set by using the image recognition model to obtain a confidence level of a recognition result corresponding to each of the multiple to-be-tagged image samples;
• select at least one to-be-tagged image sample of which the confidence level is within a preset interval as at least one to-be-tagged image sample to form a first to-be-tagged image sample set;
• acquire tags of to-be-tagged image samples in the first to-be-tagged image sample set;
• obtain an updated tagged image sample set by updating the tagged image sample set according to the acquired tags;
• and
• train the image recognition model by using the updated tagged image sample set;
• cluster the at least one to-be-tagged image sample in the first to-be-tagged image sample set to obtain a clustering result
• select part of target to-be-tagged image samples having cluster representativeness from the first to-be-tagged image sample set according to the clustering result to form a second to-be-tagged image sample set
• acquire tags of target to-be-tagged image samples in the second to-be-tagged image sample set and
• determine information entropy of a to-be-tagged image sample according to image features of the to-be-tagged image sample in the first to-be-tagged image sample set, and cluster to-be-tagged image samples of which the information entropy meets a predetermined condition to obtain K clusters and a cluster center of at least one cluster, K being a positive integer.

A non-transitory computer storage medium, having computer- executable instruction stored thereon, wherein the computer-executable instructions, when executed by one or more processors, enable the one or more processors to implement acts comprising:

• training an image recognition model by using a tagged image sample set;
• recognizing multiple to-be-tagged image samples in a to-be-tagged image sample set by using the image recognition model to obtain a confidence level of a recognition result corresponding to each of the multiple to-be-tagged image samples;
• selecting at least one to-be-tagged image sample of which the confidence level is within a preset interval to form a first to-be-tagged image sample set;
• acquiring tags of to-be-tagged image samples in the first to-be-tagged image sample set;
• obtaining an updated tagged image sample set by updating the tagged image sample set according to the acquired tags;
• and
• training the image recognition model by using the updated tagged image sample set wherein acquiring, by the terminal, the tags of the to-be-tagged image samples in the first to-be-tagged image sample set comprises:
• clustering, by the terminal device, the at least one to-be-tagged image sample in the first to-be-tagged image sample set to obtain a clustering result
• selecting, by the terminal device, part of target to-be-tagged image samples having cluster representativeness from the first to-be-tagged image sample set according to the clustering result to form a second to-be-tagged image sample set and
• acquiring, by the terminal device, tags of target to-be-tagged image samples in the second to-be-tagged image sample set and
• wherein clustering, by the terminal device, the at least one to-be-tagged image sample in the first to-be-tagged image sample set to obtain the clustering result comprises:
• determining, by the terminal device, information entropy of a to-be-tagged image sample according to image features of the to-be-tagged image sample in the first to-be-tagged image sample set and
• clustering, by the terminal device, to-be-tagged image samples of which the information entropy meets a predetermined condition to obtain K clusters and a cluster center of at least one cluster, K being a positive integer.