-
Notifications
You must be signed in to change notification settings - Fork 1
/
color_decoder.py
170 lines (136 loc) · 6.19 KB
/
color_decoder.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
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
# --------------------------------------------------------
# References:
# timm: https://github.com/rwightman/pytorch-image-models/tree/master/timm
# DeiT: https://github.com/facebookresearch/deit
# --------------------------------------------------------
import torch
import torch.nn as nn
from functools import partial
from timm.models.vision_transformer import Block
from util.pos_embed import get_2d_sincos_pos_embed
import numpy as np
class LayerNorm(nn.Module):
def __init__(self, dim):
super().__init__()
#self.ln = nn.LayerNorm((198,dim),eps=1e-6)
self.ln = nn.LayerNorm(dim,eps=1e-6)
def forward(self, x):
return self.ln(x)
class ColorDecoder(nn.Module):
""" color decoder with VisionTransformer
"""
def __init__(self, img_size=224, patch_size=16, in_chans=3,
embed_dim=1024, clip_feature_dim=512, decoder_embed_dim=512,
decoder_depth=8, decoder_num_heads=16, mlp_ratio=4., norm_layer=nn.LayerNorm):
super().__init__()
self.num_patches = (img_size // patch_size)**2
self.patch_size = patch_size
self.decoder_embed_dim = decoder_embed_dim
self.decoder_embed = nn.Linear(embed_dim, decoder_embed_dim, bias=True)
self.clip_embed = nn.Linear(clip_feature_dim, decoder_embed_dim, bias=True)
self.token_num = self.num_patches + 2
self.decoder_pos_embed = nn.Parameter(torch.zeros(1, self.token_num , decoder_embed_dim),
requires_grad=False) # fixed sin-cos embedding
self.decoder_blocks = nn.ModuleList([
Block(decoder_embed_dim, decoder_num_heads, mlp_ratio, qkv_bias=True, norm_layer=norm_layer)#remove qk_scale=None,
for i in range(decoder_depth)])
self.decoder_pred = nn.Linear(decoder_embed_dim, patch_size ** 2 * in_chans, bias=True) # decoder to patch
self.decoder_conv1 = nn.Conv2d(patch_size ** 2 * in_chans, patch_size ** 2 * in_chans, 3, stride=1, padding=(3-1)//2, bias=True)
self.color_embdding = nn.Linear(patch_size ** 2 * in_chans, decoder_embed_dim, bias=True)
self.initialize_weights()
def initialize_weights(self):
decoder_pos_embed = get_2d_sincos_pos_embed(self.decoder_pos_embed.shape[-1],
int(self.num_patches ** .5), cls_token=True)
extra_token = 1
decoder_pos_embed = np.concatenate([decoder_pos_embed, np.zeros([extra_token, self.decoder_embed_dim]) + 0.5], axis=0)
self.decoder_pos_embed.data.copy_(torch.from_numpy(decoder_pos_embed).float().unsqueeze(0))
self.apply(self._init_weights)
def _init_weights(self, m):
if isinstance(m, nn.Linear):
# we use xavier_uniform following official JAX ViT:
torch.nn.init.xavier_uniform_(m.weight)
if isinstance(m, nn.Linear) and m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.LayerNorm):
nn.init.constant_(m.bias, 0)
nn.init.constant_(m.weight, 1.0)
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.bias, 0)
nn.init.constant_(m.weight, 1.0)
elif isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
if m.bias is not None:
nn.init.constant_(m.bias, 0.)
def patchify(self, imgs):
"""
imgs: (N, 3, H, W)
x: (N, L, patch_size**2 *3)
"""
p = self.patch_size
assert imgs.shape[2] == imgs.shape[3] and imgs.shape[2] % p == 0
h = w = imgs.shape[2] // p
x = imgs.reshape(shape=(imgs.shape[0], 3, h, p, w, p))
x = torch.einsum('nchpwq->nhwpqc', x)
x = x.reshape(shape=(imgs.shape[0], h * w, p ** 2 * 3))
return x
def unpatchify(self, x):
"""
x: (N, L, patch_size**2 *3)
imgs: (N, 3, H, W)
"""
p = self.patch_size
h = w = int(x.shape[1] ** .5)
assert h * w == x.shape[1]
x = x.reshape(shape=(x.shape[0], h, w, p, p, 3))
x = torch.einsum('nhwpqc->nchpwq', x)
imgs = x.reshape(shape=(x.shape[0], 3, h * p, h * p))
return imgs
def feature_unpatchify(self,x):
p = self.patch_size
h = w = x.shape[2]
x = x.reshape(shape=(x.shape[0],p,p,3,h,w))
x = torch.einsum('npqchw->nchpwq',x)
x = x.reshape(shape=(x.shape[0],3,h*p,w*p))
return x
def forward_loss(self, pred, gray_target, target, alpha=0.):
"""
loss of pred + gray_target and traget, which have the same shape
"""
loss_l2 = (pred + gray_target - target) ** 2
loss_l2 = loss_l2.mean() # [N, L], mean loss per patch
loss_l1 = torch.abs((pred + gray_target - target))
loss_l1 = loss_l1.mean()
return loss_l2 * (1.0 - alpha) + loss_l1 * alpha
def forward(self, x, clip_x, color_mask):
# embed tokens
x = self.decoder_embed(x)
color_mask = self.patchify(color_mask)
x_color = self.color_embdding(color_mask)
x_color = torch.cat([x[:, 0, :].unsqueeze(1), x_color], dim=1)
x = x + x_color
clip_x = clip_x.unsqueeze(1)
clip_x = self.clip_embed(clip_x)
x = torch.cat([x, clip_x], dim=1)
# add pos embed
x = x + self.decoder_pos_embed
# apply Transformer blocks
for blk in self.decoder_blocks:
x = blk(x)
x = self.decoder_pred(x)
x = x[:, 1:-1, :]
h = w = int(x.shape[1] ** .5)
dim = x.shape[2]
x = x.reshape(shape=(x.shape[0], h, w, dim))
x = x.permute(0, 3, 1, 2)
x = self.decoder_conv1(x)
x = self.feature_unpatchify(x)
return x
def mae_color_decoder_base(**kwargs):
model = ColorDecoder(
patch_size=16, embed_dim=768, decoder_embed_dim=512, decoder_depth=8, decoder_num_heads=16,
mlp_ratio=4, norm_layer=LayerNorm, **kwargs)#partial(nn.LayerNorm, eps=1e-6)
return model