search

sshan-zhao / ACMNet

Adaptive Context-Aware Multi-Modal Network for Depth Completion
66 stars 11 forks source link

Source code is different from the describtion in your paper #2

Closed jikerWRN closed 4 years ago

jikerWRN commented 4 years ago

Thanks for your work! When i read your provided source code, i found that the part of decoder (SGFM) is diffierent from the decsribtion in your paper. The SGFM is described as the following figure:

1

But the code implement is different from the describtion in the figure.

    d_ufeat3 = self.d_deconv3(d_feat)  
    r_ufeat3 = self.r_deconv3(r_feat)
    d_gate3 = torch.sigmoid(self.d_gate3(d_ufeat3))
    d_feat = self.d_resblock50(torch.cat([d_feat2, d_ufeat3, d_gate3*r_feat2], 1))
    d_feat = self.d_resblock51(d_feat)
    r_gate3 = torch.sigmoid(self.r_gate3(r_ufeat3))
    r_feat = self.r_resblock50(torch.cat([r_feat2, r_ufeat3, r_gate3*d_feat2], 1))
    r_feat = self.r_resblock51(r_feat)
sshan-zhao commented 4 years ago

Could you describe it in detail? Thanks.

jikerWRN commented 4 years ago

When I write the reply, I found that I misunderstand it, please forgive my stupidity lol. May i know the reference code of "CoAttnGPBlock" part, such as knn_operation, gather_operation and grouping_operation? where is the source of pointlib? Thanks!

sshan-zhao commented 4 years ago

When I write the reply, I found that I misunderstand it, please forgive my stupidity lol. May i know the reference code of "CoAttnGPBlock" part, such as knn_operation, gather_operation and grouping_operation? where is the source of pointlib? Thanks!

https://github.com/sshan-zhao/ACMNet/tree/master/pointlib