Convolutional Neural Networks: Number of Neurons to Flatten

I made a function so you don’t have to count padding, max pooling, strides, etc when flattening a convolutional layer.

def flatten(w, k=3, s=1, p=0, z=True):
    Returns the right size of the flattened tensor after
        convolutional transformation
    :param w: width of image
    :param k: kernel size
    :param s: stride
    :param p: padding
    :param m: max pooling(bool)
    :return: proper shape and params: use x * x * previous_out_channels


    with three convolutional layers:
    r = flatten(*flatten(*flatten(w=100, k=3, s=1, p=0, m=True)))[0]

    self.fc1 = nn.Linear(r*r*128, 1024)
    return int((np.floor((w - k + 2 * p) / s) + 1) / 2 if m else 1), k, s, p, m

Here’s an example to show that it works:

import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from import DataLoader

x = np.random.rand(1_00, 3, 100, 100)
y = np.random.randint(0, 2, 1_00)

if torch.cuda.is_available():
    x = torch.from_numpy(x.astype('float32')).cuda()
    y = torch.from_numpy(y.astype('float32')).cuda()

def flatten(w, k=3, s=1, p=0, m=True):
    return int((np.floor((w - k + 2 * p) / s) + 1) / 2 if m else 1), k, s, p, m

class ConvNet(nn.Module):

    def __init__(self):
        self.conv1 = nn.Conv2d(3, 32, 3)
        self.conv2 = nn.Conv2d(32, 64, 3)
        self.conv3 = nn.Conv2d(64, 128, 3)

        r = flatten(*flatten(*flatten(w=100, k=3, s=1, p=0, m=True)))[0]
        # picture width is 100, kernel is 3, stride is 1, padding is 0, 

        self.fc1 = nn.Linear(r*r*128, 1024) # 128 is wrong here
        self.fc2 = nn.Linear(1024, 1)

    def forward(self, x):
        x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))
        x = F.max_pool2d(F.relu(self.conv2(x)), (2, 2))
        x = F.max_pool2d(F.relu(self.conv3(x)), (2, 2))
        x = x.view(x.size(0), -1)
        x = F.relu(self.fc1(x))
        x = torch.sigmoid(self.fc2(x))
        return x

net = ConvNet()
optimizer = optim.Adam(net.parameters(), lr=0.03)
loss_function = nn.BCELoss()

class Train:

    def __init__(self):
        self.len = x.shape[0]
        self.x_train = x
        self.y_train = y

    def __getitem__(self, index):
        return x[index], y[index].unsqueeze(0)

    def __len__(self):
        return self.len

train = Train()
train_loader = DataLoader(dataset=train, batch_size=64, shuffle=True)

epochs = 1
train_losses = list()
for e in range(epochs):
    running_loss = 0
    for images, labels in train_loader:
        log_ps = net(images)
        loss = loss_function(log_ps, labels)
        running_loss += loss.item()
    print('It\'s working.')

Your example doesn’t seem to contain your flatten method or at least I cannot see it. :wink:

You’re absolutely right and I am not proud of that. I put it now.

If the config of the conv layers differs in some attributes, e.g. kernel_size, how would I need to use it, as it seems I cannot overwrite k in the stacked calls like:

flatten(*flatten(w=100, k=3, s=1, p=0, m=True), k=5)

Also, the pooling operation seems to be fixed to kernel_size==stride=2 and the pooling argument also propagates?