A method of learning and testing an R-CNN-based object detector, a learning device and a test device using it.

摘要

To provide a method for learning and a method for testing an R-CNN based object detector, and a learning device and a testing device using the same.SOLUTION: A learning device 100 instructs a k-th up-sampling block to a first up-sampling block to output a (k-1)-st pyramidic feature map to a first pyramidic feature map respectively, instructs an RPN to output each ROI corresponding to each candidate region, instructs a pooling layer to output a feature vector; and learns parameters of an FC layer, the k-th up-sampling block to the first up-sampling block, and the first filter block to the n-th filter block through back-propagation of a first loss generated by referring to object class information, object regression information, and their corresponding GTs.SELECTED DRAWING: Figure 3