LEARNING METHOD AND LEARNING DEVICE FOR OBJECT DETECTOR BASED ON RECONFIGURABLE NETWORK FOR OPTIMIZING ACCORDING TO CUSTOMERS' REQUIREMENTS SUCH AS KEY PERFORMANCE INDEX USING TARGET OBJECT ESTIMATING NETWORK AND TARGET OBJECT MERGING NETWORK AND TESTING METHOD AND TESTING DEVICE USING THE SAME

摘要

A method for learning parameters of a CNN-based object detector suitable for user requirements such as a key performance index is provided using a target object prediction network and a target object integration network. The CNN may be redesigned as the scale of the object is changed by changing the resolution or focal length according to the key performance indicator. The method includes: causing the convolutional layer to output a k-th feature map by applying a convolution operation to the k-th processed image corresponding to the (k-1)-th target region on the image; and integrating the first to n-th object detection information output from the FC layer by the object integration network, and backpropagating the loss generated by referring to the integrated object detection information and the corresponding GT. . The method may be usefully performed for multi-camera, surround view monitoring, etc. since the accuracy of the 2D bounding box is improved.