Neural network model training method and device, storage medium and program product

Abstract

The invention relates to a neural network model training method and device, a storage medium and a program product. The training method of the neural network model comprises the following steps: respectively performing segmentation processing on each initial video in a plurality of initial videos to obtain a plurality of video clips corresponding to the plurality of initial videos; recombining the plurality of video clips to obtain a plurality of target videos; for each video clip, determining feature information corresponding to the video clip based on a target video to which the video clip belongs; and training a neural network model based on the feature information corresponding to each video clip. Through the method and the device, the neural network model for accurately identifying the time sequence action in the video can be obtained.

Description

technical field [0001] The present disclosure relates to the technical field of artificial intelligence, and in particular, to a training method, device, storage medium and program product of a neural network model. Background technique [0002] Video understanding tasks play a crucial role in AI applications such as action recognition and video retrieval. In systems currently on the market, the feature information of a given video can be extracted through a neural network model, and the video can be further analyzed using the extracted feature information. [0003] In the related art, the video feature learning method relies on a large number of manually annotated and labeled video data to train a neural network model. Since manual labeling of videos is a time-consuming and labor-intensive task, self-supervised learning neural network models that acquire feature representation by designing some learnable tasks or comparative learning have been proposed one after another. ...

AI Technical Summary

Problems solved by technology

However, in related technologies, neither the neural network model constructed by manual labeling nor the neural network model constructed by self-supervised learning can accurately detect and label the time-series actions that change over time in the video.

Images