CPC G06V 10/25 (2022.01) [G06T 7/70 (2017.01); G06V 10/761 (2022.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06V 2201/07 (2022.01)] | 5 Claims |
1. An image data augmentation method, comprising:
(a) training an object detection model with a high recall rate according to an image, a plurality of first bounding boxes corresponding to the image, and respective first classification labels of the plurality of first bounding boxes;
(b) using the object detection model to generate a plurality of second bounding boxes and respective second classification labels of the plurality of second bounding boxes from the image;
(c) using the plurality of first bounding boxes and the plurality of second bounding boxes as a plurality of prediction boxes, and using a plurality of first classification labels and a plurality of second classification labels as a plurality of prediction labels;
(d) identifying an overlap ratio between the two prediction boxes, and determining whether the overlap ratio between the two prediction boxes is more than a ratio threshold, wherein the two prediction boxes have the same prediction label;
(e) when the overlap ratio between the two prediction boxes is more than the ratio threshold, deleting one of the two prediction boxes to update the plurality of prediction boxes;
(f) determining whether a recursive end condition has been met according to the plurality of prediction boxes and the plurality of prediction boxes before the update;
(g) when the recursive end condition has been met, using the plurality of prediction boxes and the plurality of prediction labels for executing machine learning; and
(h) when the recursive end condition is not met, using the plurality of predicted boxes as the plurality of second bounding boxes, and using the plurality of predicted labels as a plurality of second classification labels, so as to execute steps (c) to step (f).
|