Method and apparatus for training feature extraction model, computer device, and computer-readable storage medium

#Embodiments of this application relate to the field of video processing technologies, including a method and an apparatus for training a feature extraction model, a computer device, and a computer-readable storage medium.

#With the rapid development of Internet technologies and the booming of Internet videos, video identification has been widely applied to a plurality of fields, such as video recommendation, copyright detection, target tracking, and video surveillance. Video feature extraction is a key step of video identification. To improve the accuracy of video features, usually, a feature extraction model may be first trained, and a video feature is extracted based on the feature extraction model.

#At the stage of training the feature extraction model, a plurality of images in at least one sample video are obtained, and enhancement, such as image scaling or translation is performed on the plurality of images, and a plurality of processed images are determined as sample images. Training is performed according to the determined plurality of sample images, to obtain the feature extraction model. At the stage of video identification, for a to-be-identified target video, a plurality of images in the target video are selected, and the plurality of images are inputted into the trained feature extraction model, and video features of the target video may be obtained based on the feature extraction model.

#To describe the technical solutions of the embodiments of this application more clearly, the following briefly describes the accompanying drawings required for describing the embodiments. The accompanying drawings in the following description show only some exemplary embodiments of this application, and a person of ordinary skill in the art may still derive other accompanying drawings from these accompanying drawings.

#FIG. 1A is a schematic structural diagram of an implementation environment according to an embodiment of this application.

#FIG. 1B is a flowchart of a method for training a feature extraction model according to an embodiment of this application.

#The following describes the technical solutions in exemplary embodiments of this application with reference to the accompanying drawings in the embodiments of this application. The described embodiments are some of embodiments of this application rather than all of the embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the exemplary embodiments of this application shall fall within the protection scope of this application.

#In the related art, when the feature extraction model is trained, usually, a plurality of images in at least one sample video are obtained, enhancement is performed on the plurality of images, and the processed images are used as sample images. Training is performed according to the determined plurality of sample images, to obtain the feature extraction model. By performing enhancement on the images, anti-noise performance of the feature extraction model in a spatial dimension may be improved.

#However, a video includes spatial dimension information and time dimension information. In the foregoing related art, a change of the video in a time dimension is not considered. Consequently, the feature extraction model cannot learn a change of the video in the time dimension, and has relatively poor anti-noise performance in the time dimension. When video features are extracted based on the feature extraction model, the accuracy of the video features is affected. For example, when an object in the video deforms, the feature extraction model cannot learn the deformation. To be specific, a video obtained after the video deforms cannot be detected based on the feature extraction model, and the identification accuracy is low.