Video retrieval method, device, and medium

A video retrieval method, comprising:

• extracting a keyframe of a video from an Interactive Electronic Technical Manual (IETM) common source database (CSDB) in an IETM platform,
• transferring the keyframe to a Contrastive Language-Image Pre-training (CLIP) encoder, and
• storing the keyframe in a keyframe database,
• wherein the IETM CSDB comprises a video database, the keyframe database, a text database, and a vector database;
• the CLIP encoder comprises a CLIP image encoder and a CLIP text encoder;
• an adapter is inserted into a transformer layer of the CLIP image encoder;
• and the adapter is used for parameter-efficient fine-tuning;
• using the CLIP image encoder to encode the keyframe to determine a fused vector of all keyframe features, and storing the fused vector in the vector database as a video feature;
• using the CLIP text encoder to encode video retrieval text to determine a text feature, and storing the text feature in the text database, wherein the video retrieval text is video retrieval text input by a client into the IETM platform;
• calculating a similarity between the text feature and the video feature to determine a text-video similarity matrix;
• and
• performing video retrieval based on the text-video similarity matrix, and determining a video retrieval result, wherein the IETM platform comprises a communicatively-connected display device, and controls the communicatively-connected display device to display the video retrieval text and the video retrieval result, such that an operating device performs the video retrieval based on the video retrieval text.