A method for auto-labeling a training image for use in learning a deep learning network for analyzing a high-precision image, and an auto-labeling device using the training image. IMAGES WITH HIGH PRECISION, AND AUTO-LABELING DEVICE USING THE SAME}

摘要

The present invention relates to a method for auto-labeling a training image used for learning a neural network to obtain high accuracy. In the method, (a) an auto-labeling device outputs a feature map using a meta-ROI detection network, and n current meta-ROIs in which objects on a specific training image are grouped by respective positions are displayed. And (b) the auto-labeling device crops the region corresponding to the n current meta ROIs on the training image to label n labeling images with respective bounding boxes for each of the n processing images. Outputting each processed image and merging the n labeled processed images to generate a specific labeled training image. [Selection diagram] Figure 2