Method of processing video, method of quering video, and method of training model

A method of querying a video, comprising:

• extracting a text feature of a query text;
• acquiring a target feature of each video in a plurality of candidate videos;
• and
• determining, from the plurality of candidate videos, a video matched with the query text according to the text feature and the target feature,
• wherein the target feature is obtained by using a method of processing a video, comprising operations of:
• extracting, for a video to be processed, a plurality of video features under a plurality of receptive fields;
• extracting a local feature of the video to be processed according to a video feature under a target receptive field in the plurality of receptive fields, wherein the target receptive field is any receptive field between a largest receptive field and a smallest receptive field in the plurality of receptive fields;
• obtaining a global feature of the video to be processed according to a video feature under the largest receptive field in the plurality of receptive fields;
• and
• merging the local feature and the global feature to obtain a target feature of the video to be processed,
• wherein the extracting a local feature of the video to be processed according to a video feature under a target receptive field in the plurality of receptive fields comprises:
• extracting a video feature under the target receptive field according to a sliding window with a predetermined size, so as to obtain a plurality of video sub-features;
• processing, for each sub-feature in the plurality of video sub-features, the sub-feature by using an attention algorithm, so as to obtain a processed sub-feature;
• and
• obtaining the local feature according to a plurality of processed sub-features obtained for the plurality of video sub-features.

A method of training a video processing model, wherein the video processing model comprises a video feature extraction network, a local feature extraction network and a feature merging network;

• the method comprises:
• inputting a sample video contained in sample data into the video feature extraction network, so as to obtain a plurality of video features under a plurality of receptive fields;
• inputting a video feature under a target receptive field in the plurality of receptive fields into the local feature extraction network, so as to obtain a local feature of the sample video, wherein the target receptive field is any receptive field between a largest receptive field and a smallest receptive field in the plurality of receptive fields;
• obtaining a global feature of the sample video according to a video feature under the largest receptive field in the plurality of receptive fields;
• inputting the local feature and the global feature into the feature merging network, so as to obtain a target feature of the sample video;
• acquiring a text feature of a sample text in the sample data;
• and
• training the video processing model according to the target feature and the text feature,
• wherein the inputting a video feature under a target receptive field in the plurality of receptive fields into the local feature extraction network, so as to obtain a local feature of the sample video comprises:
• by using the local feature extraction network,
• extracting a video feature under the target receptive field according to a sliding window with a predetermined size, so as to obtain a plurality of video sub-features;
• processing, for each sub-feature in the plurality of video sub-features, the sub-feature by using an attention algorithm, so as to obtain a processed sub-feature;
• and
• obtaining the local feature according to a plurality of processed sub-features obtained for the plurality of video sub-features.