Source code for catalyst.contrib.modules.pooling
import torch
import torch.nn as nn
import torch.nn.functional as F
from catalyst.contrib.registry import MODULES
[docs]class GlobalAvgPool2d(nn.Module):
def __init__(self):
super().__init__()
[docs] def forward(self, x):
h, w = x.shape[2:]
return F.avg_pool2d(input=x, kernel_size=(h, w))
[docs] @staticmethod
def out_features(in_features):
return in_features
[docs]class GlobalMaxPool2d(nn.Module):
def __init__(self):
super().__init__()
[docs] def forward(self, x):
h, w = x.shape[2:]
return F.max_pool2d(input=x, kernel_size=(h, w))
[docs] @staticmethod
def out_features(in_features):
return in_features
[docs]class GlobalConcatPool2d(nn.Module):
def __init__(self):
super().__init__()
self.avg = GlobalAvgPool2d()
self.max = GlobalMaxPool2d()
[docs] def forward(self, x):
return torch.cat([self.avg(x), self.max(x)], 1)
[docs] @staticmethod
def out_features(in_features):
return in_features * 2
[docs]class GlobalAttnPool2d(nn.Module):
def __init__(self, in_features, activation_fn="Sigmoid"):
super().__init__()
activation_fn = MODULES.get_if_str(activation_fn)
self.attn = nn.Sequential(
nn.Conv2d(
in_features, 1, kernel_size=1, stride=1, padding=0, bias=False
), activation_fn()
)
[docs] def forward(self, x):
x_a = self.attn(x)
x = x * x_a
x = torch.sum(x, dim=[-2, -1], keepdim=True)
return x
[docs] @staticmethod
def out_features(in_features):
return in_features
[docs]class GlobalAvgAttnPool2d(nn.Module):
def __init__(self, in_features, activation_fn="Sigmoid"):
super().__init__()
self.avg = GlobalAvgPool2d()
self.attn = GlobalAttnPool2d(in_features, activation_fn)
[docs] def forward(self, x):
return torch.cat([self.avg(x), self.attn(x)], 1)
[docs] @staticmethod
def out_features(in_features):
return in_features * 2
[docs]class GlobalMaxAttnPool2d(nn.Module):
def __init__(self, in_features, activation_fn="Sigmoid"):
super().__init__()
self.max = GlobalMaxPool2d()
self.attn = GlobalAttnPool2d(in_features, activation_fn)
[docs] def forward(self, x):
return torch.cat([self.max(x), self.attn(x)], 1)
[docs] @staticmethod
def out_features(in_features):
return in_features * 2
[docs]class GlobalConcatAttnPool2d(nn.Module):
def __init__(self, in_features, activation_fn="Sigmoid"):
super().__init__()
self.avg = GlobalAvgPool2d()
self.max = GlobalMaxPool2d()
self.attn = GlobalAttnPool2d(in_features, activation_fn)
[docs] def forward(self, x):
return torch.cat([self.avg(x), self.max(x), self.attn(x)], 1)
[docs] @staticmethod
def out_features(in_features):
return in_features * 3