Maxpooling not performed within each stage?

[Geevarghese George]

Hello,

Congrats on this cool work! My doubt is about why the maxpool operation takes place outside of the for loop in the forward pass of the model defined in the referenced line here:

pointMLP-pytorch/classification_ModelNet40/models/pointmlp.py

Lines 336 to 342 in 3e3d80c

 for i in range(self.stages): 
     # Give xyz[b, p, 3] and fea[b, p, d], return new_xyz[b, g, 3] and new_fea[b, g, k, d] 
     xyz, x = self.local_grouper_list[i](xyz, x.permute(0, 2, 1))  # [b,g,3]  [b,g,k,d] 
     x = self.pre_blocks_list[i](x)  # [b,d,g] 
     x = self.pos_blocks_list[i](x)  # [b,d,g] 
  
 x = F.adaptive_max_pool1d(x, 1).squeeze(dim=-1) 

Specifically, it seems that the aggregation doesnt take place, $\phi_{pos}(\phi_{pre}(f_{i,j}))$.

I would have expected to do this instead according to how its defined in the paper $\phi_{pos}(A(\phi_{pre}(f_{i,j})))$:

def forward(self, ...):
.
.
        for i in range(self.stages): 
            xyz, x = self.local_grouper_list[i](xyz, x.permute(0, 2, 1))  
            x = self.pre_blocks_list[i](x)  
            x = F.adaptive_max_pool1d(x, 1).squeeze(dim=-1)  # pooling inside the loop after pre block
            x = self.pos_blocks_list[i](x) 
.
.

Thanks for clarifying this.

[Geevarghese George]

Ah, just saw that its done within the Pre block:

pointMLP-pytorch/classification_ModelNet40/models/pointmlp.py

Line 256 in 3e3d80c

 x = F.adaptive_max_pool1d(x, 1).view(batch_size, -1) 

Thanks!