Create models/safmn_arch.py
Browse files- models/safmn_arch.py +113 -0
models/safmn_arch.py
ADDED
|
@@ -0,0 +1,113 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
import torch.nn as nn
|
| 3 |
+
import torch.nn.functional as F
|
| 4 |
+
|
| 5 |
+
|
| 6 |
+
# Layer Norm
|
| 7 |
+
class LayerNorm(nn.Module):
|
| 8 |
+
def __init__(self, normalized_shape, eps=1e-6, data_format="channels_first"):
|
| 9 |
+
super().__init__()
|
| 10 |
+
self.weight = nn.Parameter(torch.ones(normalized_shape))
|
| 11 |
+
self.bias = nn.Parameter(torch.zeros(normalized_shape))
|
| 12 |
+
self.eps = eps
|
| 13 |
+
self.data_format = data_format
|
| 14 |
+
if self.data_format not in ["channels_last", "channels_first"]:
|
| 15 |
+
raise NotImplementedError
|
| 16 |
+
self.normalized_shape = (normalized_shape, )
|
| 17 |
+
|
| 18 |
+
def forward(self, x):
|
| 19 |
+
if self.data_format == "channels_last":
|
| 20 |
+
return F.layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps)
|
| 21 |
+
elif self.data_format == "channels_first":
|
| 22 |
+
u = x.mean(1, keepdim=True)
|
| 23 |
+
s = (x - u).pow(2).mean(1, keepdim=True)
|
| 24 |
+
x = (x - u) / torch.sqrt(s + self.eps)
|
| 25 |
+
x = self.weight[:, None, None] * x + self.bias[:, None, None]
|
| 26 |
+
return x
|
| 27 |
+
|
| 28 |
+
# CCM
|
| 29 |
+
class CCM(nn.Module):
|
| 30 |
+
def __init__(self, dim, growth_rate=2.0):
|
| 31 |
+
super().__init__()
|
| 32 |
+
hidden_dim = int(dim * growth_rate)
|
| 33 |
+
|
| 34 |
+
self.ccm = nn.Sequential(
|
| 35 |
+
nn.Conv2d(dim, hidden_dim, 3, 1, 1),
|
| 36 |
+
nn.GELU(),
|
| 37 |
+
nn.Conv2d(hidden_dim, dim, 1, 1, 0)
|
| 38 |
+
)
|
| 39 |
+
|
| 40 |
+
def forward(self, x):
|
| 41 |
+
return self.ccm(x)
|
| 42 |
+
|
| 43 |
+
|
| 44 |
+
# SAFM
|
| 45 |
+
class SAFM(nn.Module):
|
| 46 |
+
def __init__(self, dim, n_levels=4):
|
| 47 |
+
super().__init__()
|
| 48 |
+
self.n_levels = n_levels
|
| 49 |
+
chunk_dim = dim // n_levels
|
| 50 |
+
|
| 51 |
+
# Spatial Weighting
|
| 52 |
+
self.mfr = nn.ModuleList([nn.Conv2d(chunk_dim, chunk_dim, 3, 1, 1, groups=chunk_dim) for i in range(self.n_levels)])
|
| 53 |
+
|
| 54 |
+
# # Feature Aggregation
|
| 55 |
+
self.aggr = nn.Conv2d(dim, dim, 1, 1, 0)
|
| 56 |
+
|
| 57 |
+
# Activation
|
| 58 |
+
self.act = nn.GELU()
|
| 59 |
+
|
| 60 |
+
def forward(self, x):
|
| 61 |
+
h, w = x.size()[-2:]
|
| 62 |
+
|
| 63 |
+
xc = x.chunk(self.n_levels, dim=1)
|
| 64 |
+
out = []
|
| 65 |
+
for i in range(self.n_levels):
|
| 66 |
+
if i > 0:
|
| 67 |
+
p_size = (h//2**i, w//2**i)
|
| 68 |
+
s = F.adaptive_max_pool2d(xc[i], p_size)
|
| 69 |
+
s = self.mfr[i](s)
|
| 70 |
+
s = F.interpolate(s, size=(h, w), mode='nearest')
|
| 71 |
+
else:
|
| 72 |
+
s = self.mfr[i](xc[i])
|
| 73 |
+
out.append(s)
|
| 74 |
+
|
| 75 |
+
out = self.aggr(torch.cat(out, dim=1))
|
| 76 |
+
out = self.act(out) * x
|
| 77 |
+
return out
|
| 78 |
+
|
| 79 |
+
class AttBlock(nn.Module):
|
| 80 |
+
def __init__(self, dim, ffn_scale=2.0):
|
| 81 |
+
super().__init__()
|
| 82 |
+
|
| 83 |
+
self.norm1 = LayerNorm(dim)
|
| 84 |
+
self.norm2 = LayerNorm(dim)
|
| 85 |
+
|
| 86 |
+
# Multiscale Block
|
| 87 |
+
self.safm = SAFM(dim)
|
| 88 |
+
# Feedforward layer
|
| 89 |
+
self.ccm = CCM(dim, ffn_scale)
|
| 90 |
+
|
| 91 |
+
def forward(self, x):
|
| 92 |
+
x = self.safm(self.norm1(x)) + x
|
| 93 |
+
x = self.ccm(self.norm2(x)) + x
|
| 94 |
+
return x
|
| 95 |
+
|
| 96 |
+
|
| 97 |
+
class SAFMN(nn.Module):
|
| 98 |
+
def __init__(self, dim, n_blocks=8, ffn_scale=2.0, upscaling_factor=4):
|
| 99 |
+
super().__init__()
|
| 100 |
+
self.to_feat = nn.Conv2d(3, dim, 3, 1, 1)
|
| 101 |
+
|
| 102 |
+
self.feats = nn.Sequential(*[AttBlock(dim, ffn_scale) for _ in range(n_blocks)])
|
| 103 |
+
|
| 104 |
+
self.to_img = nn.Sequential(
|
| 105 |
+
nn.Conv2d(dim, 3 * upscaling_factor**2, 3, 1, 1),
|
| 106 |
+
nn.PixelShuffle(upscaling_factor)
|
| 107 |
+
)
|
| 108 |
+
|
| 109 |
+
def forward(self, x):
|
| 110 |
+
x = self.to_feat(x)
|
| 111 |
+
x = self.feats(x) + x
|
| 112 |
+
x = self.to_img(x)
|
| 113 |
+
return x
|