-
Notifications
You must be signed in to change notification settings - Fork 0
/
MNIST_skeleton.py
65 lines (52 loc) · 1.73 KB
/
MNIST_skeleton.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
torch.manual_seed(4242)
data_dir = "./data/"
train_loader = torch.utils.data.DataLoader(
datasets.MNIST(data_dir, train=True, download=True,
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.30801,))
])),
batch_size=64, shuffle=True)
# TODO: implement squash function
def squash(input):
return input
conv1_params = {
"in_channels": 1,
"out_channels": 256,
"kernel_size": 9,
"stride": 1
}
conv2_params = {
"in_channels": 256,
"out_channels": 256,
"kernel_size": 9,
"stride": 2
}
class PrimaryCapsules(nn.Module):
def __init__(self, conv1_params, conv2_params, caps_maps=32, caps_dims=8):
super(PrimaryCapsules, self).__init__()
self.caps_maps = caps_maps
self.n_caps = caps_maps * 6 * 6
self.cap_dims = caps_dims
self.conv1 = nn.Conv2d(**conv1_params)
self.conv2 = nn.Conv2d(**conv2_params)
def forward(self, x):
out1 = F.relu(self.conv1(x))
print(f"Output size 1: {out1.size()}")
out2 = F.relu(self.conv2(out1))
print(f"Output size 2: {out2.size()}")
out3 = out2.view(x.size(0), -1, self.cap_dims)
# Not sure of out3 dims. May be backwards.
print(f"Output size 3: {out3.size()}")
return squash(out3)
model = PrimaryCapsules(conv1_params, conv2_params)
for batch_idx, (data, target) in enumerate(train_loader):
test_sample = data[0, :, :, :]
print(f"Sample size: {test_sample.size()}")
output = model(data)
break