davrot/kk_contour_net_shallow
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Add files via upload

Browse source
This commit is contained in:
David Rotermund 2023-07-23 01:45:45 +02:00 committed by GitHub
parent fc782baeb3
commit b8000f87a5
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 18 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
functions/SoftmaxPower.py
View file

@ -4,11 +4,15 @@ import torch
class SoftmaxPower(torch.nn.Module): class SoftmaxPower(torch.nn.Module):
dim: int | None dim: int | None
power: float power: float
mean_mode: bool
def __init__(self, power: float = 2.0, dim: int | None = None) -> None: def __init__(
self, power: float = 2.0, dim: int | None = None, mean_mode: bool = False
) -> None:
super().__init__() super().__init__()
self.dim = dim self.dim = dim
self.power = power self.power = power
self.mean_mode = mean_mode
def __setstate__(self, state): def __setstate__(self, state):
super().__setstate__(state) super().__setstate__(state)
@ -16,6 +20,8 @@ class SoftmaxPower(torch.nn.Module):
self.dim = None self.dim = None
if not hasattr(self, "power"): if not hasattr(self, "power"):
self.power = 2.0 self.power = 2.0
if not hasattr(self, "mean_mode"):
self.mean_mode = False
def forward(self, input: torch.Tensor) -> torch.Tensor: def forward(self, input: torch.Tensor) -> torch.Tensor:
output: torch.Tensor = torch.abs(input).pow(self.power) output: torch.Tensor = torch.abs(input).pow(self.power)
@ -23,7 +29,11 @@ class SoftmaxPower(torch.nn.Module):
output = output / output.sum() output = output / output.sum()
else: else:
output = output / output.sum(dim=self.dim, keepdim=True) output = output / output.sum(dim=self.dim, keepdim=True)
return input * output
if self.mean_mode:
return torch.abs(input).mean(dim=1, keepdim=True) * output
else:
return input * output
def extra_repr(self) -> str: def extra_repr(self) -> str:
return f"dim={self.dim} ; power={self.power}" return f"dim={self.dim} ; power={self.power}"

11
functions/make_cnn.py
View file

@ -16,6 +16,7 @@ def make_cnn(
pooling_type: str, pooling_type: str,
conv_0_enable_softmax: bool, conv_0_enable_softmax: bool,
conv_0_power_softmax: float, conv_0_power_softmax: float,
conv_0_meanmode_softmax: bool,
l_relu_negative_slope: float, l_relu_negative_slope: float,
) -> torch.nn.Sequential: ) -> torch.nn.Sequential:
assert len(conv_out_channels_list) >= 1 assert len(conv_out_channels_list) >= 1
@ -60,11 +61,11 @@ def make_cnn(
assert setting_understood assert setting_understood
if conv_0_enable_softmax: if conv_0_enable_softmax:
# if conv_0_power_softmax != 0.0: cnn.append(
# cnn.append(SoftmaxPower(dim=1, power=conv_0_power_softmax)) SoftmaxPower(
# else: dim=1, power=conv_0_power_softmax, mean_mode=conv_0_meanmode_softmax
# cnn.append(torch.nn.Softmax(dim=1)) )
cnn.append(SoftmaxPower(dim=1, power=conv_0_power_softmax)) )
# Changing structure # Changing structure
for i in range(1, len(conv_out_channels_list)): for i in range(1, len(conv_out_channels_list)):