BSDCNBounded Static Deformable Convolution Network LEARNING METHOD AND LEARNING DEVICE FOR BSDCNBOUNDED STATIC DEFORMABLE CONVOLUTION NETWORK DESIGNED BY OPTIMIZING DEFORMABLE CONVOLUTION NETWORK USING STATIC CALCULATION SCHEME AND TESTING METHOD AND TESTING DEVICE USING THE SAME

摘要

By introducing the concept of Deformable Convolution to increase the modeling ability for unstructured patterns, while using static calculation techniques, it can be implemented through the high-level API of general deep learning frameworks, thereby reducing the amount of computation and reducing the complexity of implementation. A learning method of an optimized BSDCN (Bounded Static Deformable Convolution Network) is disclosed. That is, (a) the learning device causes the deforming unit included in the BSDCN to cause (i) at least one original convolutional kernel and (ii) each original of the original convolutional kernel when a training image is input. generating at least one deformed convolutional kernel that is expanded and distributed compared to the original convolutional kernel with reference to at least one offset parameter corresponding to at least some of the convolutional parameters; (b) causing, by the learning device, the deformed convolutional layer included in the BSDCN to generate at least one intermediate feature map for learning by applying at least one deformed convolution operation to the training image using the deformed convolution kernel. ; (c) causing, by the learning device, the computation layer included in the BSDCN to generate inference information for learning corresponding to the learning image by applying at least one neural network operation to the intermediate feature map for learning; and (d) the learning device causes the loss layer included in the BSDCN to generate a loss with reference to the inference information for learning and the Ground-Truth inference information, and then perform backpropagation with reference to the loss learning at least a portion of the offset parameter and the original convolutional parameter by performing