The difference between the code and the original paper.

[zhouyuan888888]

Hi, thank you for releasing the code for EMANet. I find a difference between the code and the paper. The difference lies in the formulation of Equation 13 (in the paper). In the paper, the M step (bases reconstruct) is formulated as follows: image However, in the code, the M step is formulated as: mu = torch.bmm(x, z)
Actually， mu = torch.bmm(x, z) is the weighted summation of X. However, Equation 13 (in the paper) is not the weighted summation of X. Anything wrong in the paper?

[zhouyuan888888]

In the paper, M step is formulated as: image in which Znk denotes the n-th pixel in the K-th channel in Z (Attention map), Xn denotes the feature vector of the n-th pixel. Why uk is obtained from the multiplication between Znk (scalar) and Xn? Expect for your explanations. Thank you so much.

[XiaLiPKU]

Hi, thank you for releasing the code for EMANet. I find a difference between the code and the paper. The difference lies in the formulation of Equation 13 (in the paper). In the paper, the M step (bases reconstruct) is formulated as follows: image However, in the code, the M step is formulated as: mu = torch.bmm(x, z) Actually， mu = torch.bmm(x, z) is the weighted summation of X. However, Equation 13 (in the paper) is not the weighted summation of X. Anything wrong in the paper?

To be honest, I forgot the \sum symbol in the numerator... When I found that, the deadline of the camera-ready version had gone...

[zhouyuan888888]

@XiaLiPKU hahaha....， thank you for your explanation.