Need help with 3d classifier

vision
1

I can’t understand how to change this code to convert it from the classifier of the whole 3d grid to a classifier of every cell in a grid.
Initial dimensionality of every grid is 16 x 12 x 10 and 2 classes for every cell

class Grids_Conv3d(nn.Module):
    def __init__(self, out_1=32, out_2=64):
        super(Grids_Conv3d, self).__init__()
        self.layer1 = nn.Sequential(
            nn.Conv3d(in_channels=1, out_channels=out_1, kernel_size=5, stride=1, padding=2),
            nn.ReLU(),
            nn.MaxPool3d(kernel_size=2, stride=2))
        self.layer2 = nn.Sequential(
            nn.Conv3d(in_channels=out_1, out_channels=out_2, kernel_size=5, stride=1, padding=2),
            nn.ReLU(),
            nn.MaxPool3d(kernel_size=2, stride=2))
        self.drop_out = nn.Dropout()
        self.fc1 = nn.Linear(1536, 1920)
        self.fc2 = nn.Linear(1920, 2) #number of classes

def forward(self, x):
    out = self.layer1(x)
    out = self.layer2(out)
    out = out.view(out.size(0), -1)
    out = self.drop_out(out)
    out = self.fc1(out)
    out = self.fc2(out)
    return F.log_softmax(out, dim=1)

Could anyone help me with this?
Thank you!

2

What does “cell” mean in this context?
Based on your description, I assume you would like to index smaller “cells” in your grid and feed each one into your model.
If that’s the case, you would have to make sure the output size of layer2 stays the same, so that the in_features of self.fc1 don’t need to be changed.

3

Thanx for your reply.
Every grid is an 3D object wich consists of cells. Every cell is a property and cells have some spatial correlation which I want to reproduce by cells classification.
So I’m not sure (or understand how) that indexing of cells would help me.