Recently, I was looking at the paper entitled Convolutional neural network architecture for geometric matching. However, when I implemented the neural network mentioned in the article, I encountered a problem:How do you generate a random affine matrix? When I used your generated matrices, I trained 47% of the PCK on the network, but when I tried to use my randomly generated matrix, PCK became 27% and I could not believe it I really want to know how you guys did it.
Can anyone answer me？
Recently, I was looking at the paper entitled Convolutional neural network architecture for geometric matching. However, when I implemented the neural network mentioned in the article, I encountered a problem:How do you generate a random affine matrix? When I used your generated matrices, I trained 47% of the PCK on the network, but when I tried to use my randomly generated matrix, PCK became 27% and I could not believe it I really want to know how you guys did it. Can anyone answer me？