Model training method and device for multi-task knowledge fusion

• Main Authors: ZHANG AIJIA, QIU YUNLEI, DUAN MANNI, WANG YUZHU
• Format: Patent
• Language: Chinese; English
• Published: 09.05.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention discloses a model training method and device for multi-task knowledge fusion. First, image data is acquired. Secondly, according to the number of feature blocks in a to-be-trained target recognition model, performing identical transformation on each pre-trained recognition model to obtain each equivalent feature block recognition model; then, aiming at each feature block contained in the target recognition model, determining a feature block corresponding to the feature block from each equivalent feature block recognition model as a target feature block, splicing image features output by each target feature block to obtain a spliced image feature corresponding to the feature block, and carrying out feature extraction on the spliced image feature; and taking the deviation between the image feature output by the feature block and the spliced image feature corresponding to the feature block as the deviation corresponding to the feature block. And finally, taking minimization of deviation correspondi