首页>
外国专利>
CNN LEARNING METHOD AND LEARNING DEVICE FOR EXTRACTING FEATURE FROM INPUT IMAGE BY USING CONVOLUTIONAL LAYERS IN MULTIPLE BLOCKS IN CNN RESULTING IN HARDWARE OPTIMIZATION WHICH ALLOWS KEY PERFORMANCE INDEX TO BE SATISFIED AND TESTING METHOD AND TESTINGDEVICE USING THE SAME
CNN LEARNING METHOD AND LEARNING DEVICE FOR EXTRACTING FEATURE FROM INPUT IMAGE BY USING CONVOLUTIONAL LAYERS IN MULTIPLE BLOCKS IN CNN RESULTING IN HARDWARE OPTIMIZATION WHICH ALLOWS KEY PERFORMANCE INDEX TO BE SATISFIED AND TESTING METHOD AND TESTINGDEVICE USING THE SAME
In the present invention, in the method of extracting features from the input image using the first to n-th blocks in the CNN, the learning apparatus causes the first convolutional layer of the k-th block to be a 1_1 to a k_1 feature map or a From each of the predetermined feature maps summed by element, and causing the second convolutional layer of the k-th block to generate the k_2 feature map; Causing the pooling layer to input a pooled feature map generated by pooling an ROI region on an n_2 feature map output from the n-th block or a predetermined feature map calculated therefrom, as a feature classifier; It provides a method characterized in that it comprises; a step of causing the loss layer to calculate the loss by referring to the output value of the feature classifier and the corresponding GT, and optimize the hardware to improve the CNN throughput and the present invention The test method can be suitably used in small networks, mobile devices, etc., and can satisfy the Key Performance Index (KPI).
展开▼
