Data labeling method based on hard example mining and apparatus, computing device, and storage medium

A method for updating a hard example mining algorithm, wherein the method is applied to a computer device and comprises:

• determining a plurality of hard examples in an unlabeled image set by using the hard example mining algorithm;
• wherein the plurality of hard examples and hard example attributes are determined based on an inference result obtained by inputting the unlabeled image set to an initial artificial intelligence (AI) model or a trained AI model,
• displaying at least one hard example in the plurality of hard examples to a user by using a display interface;
• and
• updating the hard example mining algorithm based on a labeling result obtained after the user confirms the at least one hard example by using the display interface based on hard example coefficients of the plurality of hard examples, wherein a hard example coefficient reflects a hard example degree of the hard example, and the hard example degree is associated with a difficulty of obtaining a correct result by performing classification or detection by using the initial AI model or the trained AI model;
• wherein the method further comprises:
• determining hard example reasons of the plurality of hard examples by using the hard example mining algorithm, and providing suggestion information corresponding to the hard example reasons for the user by using the display interface, wherein the hard example reason includes a plurality of types, which are classified into data enhancement inconsistency, data feature distribution inconsistency, and
• wherein the suggestion information indicates a processing manner to be performed to reduce hard examples with corresponding hard example reasons.

A computing device, comprising:

• a memory and a processor, wherein the memory is configured to store a group of computer instructions;
• and
• the processor which is coupled to the memory executes the group of computer instructions to:
• determine, by hard example mining, a plurality of hard examples in an unlabeled image set and hard example attributes of the plurality of hard examples, wherein the plurality of hard examples and the hard example attributes are determined based on an inference result obtained by inputting the unlabeled image set to an initial artificial intelligence (AI) model or a trained AI model, and the hard example attribute comprises a hard example coefficient which reflects a hard example degree of the hard example, and the hard example degree is associated with a difficulty of obtaining a correct result by performing classification or detection by using the initial AI model or the trained AI model;
• display at least one hard example in the plurality of hard examples and a corresponding hard example attribute to a user by using a display interface;
• and
• obtain a labeling result obtained after the user confirms the at least one hard example in the display interface based on hard example coefficients of the plurality of hard examples;
• wherein the hard example attribute further comprises a hard example reason;
• and wherein the displaying the at least one hard example in the plurality of hard examples and the corresponding hard example attribute by using the display interface further comprises:
• displaying, by using the display interface, a hard example reason corresponding to the at least one hard example and suggestion information corresponding to the hard example reason, wherein
• the hard example reason includes a plurality of types, which are classified into data enhancement inconsistency, data feature distribution inconsistency, and
• the suggestion information indicates a processing manner to be performed to reduce hard examples with corresponding hard example reasons.