1xH CNN LEARNING METHOD AND LEARNING DEVICE FOR CONVOLUTIONAL NEURAL NETWORK USING 1xH CONVOLUTION FOR IMAGE RECOGNITION TO BE USED FOR HARDWARE OPTIMIZATION AND TESTING METHOD AND TESTING DEVICE USING THE SAME

摘要

The present invention provides a method for learning a CNN parameter for image recognition provided to be used for optimizing hardware that meets a KPI (Key Performance Index, key performance indicator), the learning apparatus comprising: (a) a first transposing causing the layer or the pooling layer to concatenate pixels on the pooled feature map for each ROI to generate an integrated feature map; (b) cause the 1xH1 convolutional layer to generate a first adjusted feature map using the first reshaped feature map generated by concatenating the features in H1 channels of the integrated feature map, and cause the 1xH2 convolutional layer to generating a second steering feature map using a second reshaped feature map generated by concatenating features in H2 channels of the first steering feature map; and (c) causing the second transposing layer or classification layer to separate the second adjustment feature map for each pixel to generate a feature map for each pixel; It is characterized in that it includes.