Model training and target detection method and device, storage medium and electronic equipment

Abstract

The invention discloses a model training method and device, a target detection method and device, a storage medium and electronic equipment. When the corresponding relationship between the marking frame and the reference frame is established, a part of reference frames closer to the marking frame can be screened for the marking frame according to the positions of the marking frame and the reference frame, and the corresponding relationship between the marking frame and the screened reference frame is only established for the marking frame. The calculation resources consumed for determining the positions of the labeling frame and the reference frame and determining the distance between the labeling frame and the reference frame are smaller than the calculation resources consumed for calculating the overlap ratio between the labeling frame and the reference frame. By adopting the model training method provided by the specification, the corresponding relationship between the labeling box and the reference box can be established more quickly under the condition of occupying less computing resources.

Description

technical field [0001] This specification relates to the field of computer technology, in particular to methods, devices, storage media and electronic equipment for model training and target detection. Background technique [0002] At present, through some object detection models, the algorithm based on the reference frame can be used to achieve object detection. [0003] In the idea of ​​the target detection algorithm based on the reference frame, the detection frame can be predicted based on the reference frame anchored in advance, for example, the detection frame can be located by determining the position offset based on the reference frame. [0004] During the training process of the target detection model, a label frame is usually marked in the sample image to be input into the model, and several reference frames are generated. According to the needs of the target detection task, the label of the annotation box can include the position and size of the annotation box, a...

AI Technical Summary

Problems solved by technology

[0006] However, when determining the corresponding relationship between the reference frame and the label frame, for each label frame, it is necessary to calculate the degree of coincidence between the label frame and all reference frames, so that the reference frame with a greater degree of coincidence with the label frame The corresponding relationship between the frame and the label frame, and the calculation of the coincidence degree will consume a lot of computing resources

Images