True value box selection method and device in target detection, storage medium and equipment

Abstract

The embodiment of the invention discloses a truth value box selection method and device in target detection, a storage medium and equipment and belongs to the technical field of image processing. Themethod comprises steps of obtaining a target feature map obtained after feature extraction is performed on an image, wherein the target feature map comprises a plurality of grids with predetermined sizes; obtaining a plurality of detection boxes corresponding to each small target object in the image from the target feature map; for each detection frame, calculating a centrality score of a grid ofwhich a predetermined point is located in the detection frame, the predetermined point being an angular point and / or a central point of the grid; and for each small target object, determining the detection box corresponding to the maximum centrality score as a truth value box of the small target object from a plurality of detection boxes corresponding to the small target object. According to the method, accuracy of truth value box selection can be improved.

Description

technical field [0001] The embodiments of the present application relate to the technical field of image processing, and in particular to a method, device, storage medium and equipment for selecting a truth frame in object detection. Background technique [0002] The FCOS (Fully Convolutional One-Stage Object Detection) algorithm realizes anchor-free target detection. When using the FCOS algorithm to detect the target object in the image, the image can be divided into S×S grids, multiple detection frames containing the target object can be predicted, and the detection frame can be selected according to the positional relationship between the detection frame and the grid. The ground-truth box is retained as a positive sample, and other detection boxes are discarded as negative samples to identify the target object. [0003] When using the FCOS algorithm to detect small target objects, such as figure 1 As shown, the detection frame 1 contains the small target object to be re...

AI Technical Summary

Problems solved by technology

like figure 2 As shown, the detection frame 2 contains the small target object to be regressed. Although the detection frame 2 is in nine grids, only the center point of one grid falls into the detection frame 2, so that the center of the grid used for regression fewer points

[0004] like figure 1 and 2 As shown in , even if the detection frame contains a small target object, it may be because the center point of the grid does not fall in the detection frame, or the number of grid center points falling in the detection frame is less, which affects the selection of the ground truth frame accuracy

Images