Heritage Science

Table 3 Comparison of several common neural network models

From: Ancient mural segmentation based on a deep separable convolution network

Model	Parameters	Size	Depth
Xception	22,910,480	88 MB	126
VGG19	143,667,240	549 MB	26
ResNet50	25,636,710	99 MB	168
InceptionV3	23,851,734	92 MB	159
MobileNetV2	3,538,984	14 MB	88

Parameter and size embody the complexity of the model, and high values indicate higher complexity; depth represents the depth of the training model, and a higher value indicates longer time consumption

Back to article page