Video description data processing method and device and storage medium

Abstract

The invention provides a video description data processing method and device and a storage medium, and belongs to the technical field of image processing, and the method comprises the steps: S1, importing video data, and constructing an encoder and a plurality of sequentially arranged LSTM (Long Short-Term Memory) networks; s2, encoding the video data through an encoder to obtain a visual feature vector; s3, real word vectors are imported, the LSTM long short-term memory networks, the visual feature vectors and the real word vectors are taken as one group, and each group of LSTM long short-term memory networks is sequentially judged and analyzed to obtain video description information; s4, performing loss analysis on the video description information to obtain a target video description model; and S5, importing to-be-tested video data, and performing video description on the to-be-tested video data through the target video description model to obtain a video description result. According to the method, the problem of inconsistent data distribution is solved, the generated words can be closer to real tags, and the accuracy of description generation is further improved.

Description

Technical field [0001] The present invention relates primarily to the field of image processing technology, specifically to a video description data processing method, apparatus and storage medium. Background [0002] At present, the basic training model will only use the real label as the data input when training, and the generated value can only be used as the input when testing, which causes a mismatch between training and testing. At the same time, the description and real labels generated by the existing basic model are often not appropriate enough, which reduces the accuracy of the model. Contents of the Invention [0003] The technical problem to be solved in the present invention is to provide a video description of the data processing method, apparatus and storage medium for the deficiencies of prior art. [0004] The present invention solves the technical solution of the above technical problems as follows: a video description of a data processing method, comprising t...

AI Technical Summary

Problems solved by technology

[0002] At present, the basic training model only uses real labels as data input during training, and only generates values ​​as input during testing, which causes a mismatch between training and testing.

At the same time, the descriptions and real labels generated by the existing basic models are often not appropriate enough, which reduces the accuracy of the model

Images