Method, computer device, and storage medium, for feature fusion model training and sample retrieval

A computer device, comprising a memory and a processor, the memory storing computer-readable instructions, and the processor, when executing the computer-readable instructions, implementing a method for feature fusion model training, the method comprising:

• acquiring a training sample and a corresponding training sample category label, and inputting the training sample into an initial feature fusion model;
• extracting a feature of the training sample through the initial feature fusion model to obtain a training semantic feature and a training global feature, and performing classification and recognition based on the training semantic feature to obtain an initial training category corresponding to the training sample;
• splicing the training semantic feature and the training global feature through the initial feature fusion model to obtain a spliced training feature, calculating an autocorrelation feature based on the spliced training feature to obtain an autocorrelation feature, calculating a self-attention weight based on the autocorrelation feature to obtain a self-attention weight, and adjusting the spliced training feature through the self-attention weight to obtain a fused training feature;
• calculating a feature loss based on the training global feature, the training semantic feature, and the fused training feature to obtain an initial feature loss, and calculating a category loss based on the initial training category and the training sample category label to obtain an initial category loss;
• and
• updating the initial feature fusion model based on the initial feature loss and the initial category loss, returning the acquired training sample and the corresponding training sample category label, and iteratively performing the operation of inputting the training sample into the initial feature fusion model until a target fusion model is obtained when a training completion condition is satisfied, the target fusion model being configured to fuse the semantic feature and the global feature extracted from input data to obtain a target fused feature, and use the target fused feature as an index of the input data.

A non-transitory computer-readable storage medium, storing computer-readable instructions, the computer-readable instructions, when executed by a processor, implementing the operations of a method for feature fusion model training, the method comprising:

• acquiring a training sample and a corresponding training sample category label, and inputting the training sample into an initial feature fusion model;
• extracting a feature of the training sample through the initial feature fusion model to obtain a training semantic feature and a training global feature, and performing classification and recognition based on the training semantic feature to obtain an initial training category corresponding to the training sample;
• splicing the training semantic feature and the training global feature through the initial feature fusion model to obtain a spliced training feature, calculating an autocorrelation feature based on the spliced training feature to obtain an autocorrelation feature, calculating a self-attention weight based on the autocorrelation feature to obtain a self-attention weight, and adjusting the spliced training feature through the self-attention weight to obtain a fused training feature;
• calculating a feature loss based on the training global feature, the training semantic feature, and the fused training feature to obtain an initial feature loss, and calculating a category loss based on the initial training category and the training sample category label to obtain an initial category loss;
• and
• updating the initial feature fusion model based on the initial feature loss and the initial category loss, returning the acquired training sample and the corresponding training sample category label, and iteratively performing the operation of inputting the training sample into the initial feature fusion model until a target fusion model is obtained when a training completion condition is satisfied, the target fusion model being configured to fuse the semantic feature and the global feature extracted from input data to obtain a target fused feature, and use the target fused feature as an index of the input data.