-
Notifications
You must be signed in to change notification settings - Fork 3
/
exp_face_parsing_unrefactored.py
157 lines (133 loc) · 5.56 KB
/
exp_face_parsing_unrefactored.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
import os
import shutil
import pickle
import torch
import torch.nn as nn
import fire
from tqdm import tqdm
from sklearn.cluster import KMeans
from segmentation_in_style.models.stylegan2.model import Generator
from src.models.dife import DIFE
def get_generator(type):
if type == "human":
g_ema = Generator(1024, 512, 8)
g_ema.load_state_dict(torch.load('./checkpoints/stylegan_seg/stylegan2-ffhq-config-f.pt')["g_ema"], strict=False)
g_ema.eval()
g_ema = g_ema.cuda()
elif type == "dog":
g_ema = Generator(512, 512, 8)
g_ema.load_state_dict(torch.load('./checkpoints/stylegan_seg/animals_dog_ada.pth'), strict=True)
g_ema.eval()
g_ema = g_ema.cuda()
elif type == "cat":
g_ema = Generator(512, 512, 8)
g_ema.load_state_dict(torch.load('./checkpoints/stylegan_seg/animals_cat_ada.pth'), strict=True)
g_ema.eval()
g_ema = g_ema.cuda()
elif type == "wild":
g_ema = Generator(512, 512, 8)
g_ema.load_state_dict(torch.load('./checkpoints/stylegan_seg/animals_wild_ada.pth'), strict=True)
g_ema.eval()
g_ema = g_ema.cuda()
else:
raise ValueError(f"type is wrong, type: {type}")
return g_ema
def main(n_colors=9):
resume_path = "outputs/wflw+animalweb/human+dog/003/model_best_iter_0012000.pth"
data_type = "human+dog"
count = 100
image_size = 96
save_dir = f"output_exp/exp_face_parsing/{data_type}_{n_colors}"
if os.path.exists(save_dir):
shutil.rmtree(save_dir)
os.makedirs(save_dir, exist_ok=True)
net = DIFE(16, 512, 2)
net.resume(resume_path)
net.cuda()
net.eval()
source, target = data_type.split('+')
sg2_source = get_generator(source)
sg2_target = get_generator(target)
mean_latent_source = sg2_source.mean_latent(4096)
mean_latent_target = sg2_target.mean_latent(4096)
truncation = 0.7
features = None
imgs = None
with torch.no_grad():
for i in tqdm(range(count)):
if i < count // 2:
sample_z = torch.randn(1, 512).cuda()
style = sg2_source.style(sample_z)
style = mean_latent_source + truncation * (style - mean_latent_source)
img, _, _ = sg2_source([style], input_is_latent=True, randomize_noise=False, feature_layer_number_to_return=7)
else:
sample_z = torch.randn(1, 512).cuda()
style = sg2_target.style(sample_z)
style = mean_latent_target + truncation * (style - mean_latent_target)
img, _, _ = sg2_target([style], input_is_latent=True, randomize_noise=False, feature_layer_number_to_return=7)
img = nn.functional.upsample(
img,
size=(image_size, image_size),
mode='bilinear',
align_corners=True,
).clamp(min=-1.0, max=1.0).detach()
out = net.forward_hg(img)
out = nn.functional.upsample(
out,
size=(image_size, image_size),
mode='bilinear',
align_corners=True,
).detach()
if imgs is None:
imgs = img.cpu()
else:
aditional_imgs = img.cpu()
imgs = torch.cat((imgs, aditional_imgs), axis=0)
if features is None:
features = out.cpu()
else:
additional_features = out.cpu()
features = torch.cat((features, additional_features), axis=0)
print(f"[INFO] features: {features.shape}")
_, C, _, _ = features.shape
features_new = features.permute(0, 2, 3, 1).reshape(-1, C)
arr = features_new.detach().cpu().numpy()#dist.detach().cpu().numpy().reshape(-1, 1)
kmeans = KMeans(n_clusters=n_colors, random_state=903).fit(arr)
with open(f"{save_dir}/kmeans_cluster.pkl", "wb") as f:
pickle.dump(kmeans, f)
labels = kmeans.labels_
labels_spatial = labels.reshape(features.shape[0], features.shape[2], features.shape[3])
from src.utils.visualize import Visualizer
for i in tqdm(range(count // 2)):
image_path = f"{save_dir}/{i:06d}.png"
vis = Visualizer(
save_path=image_path,
grid=(3, 2),
)
vis.draw(imgs[i], "image_tensor_chw", 1, 1)
vis.draw(imgs[i + count // 2], "image_tensor_chw", 1, 2)
vis.draw(features[i], "embedding_tensor_chw", 2, 1)
vis.draw(features[i + count // 2], "embedding_tensor_chw", 2, 2)
vis.draw(labels_spatial[i], "cvimage", 3, 1)
vis.draw(labels_spatial[i + count // 2], "cvimage", 3, 2)
vis.save()
vis = Visualizer(save_path=f"{save_dir}/{i:06d}_1_src_img.png")
vis.draw(imgs[i], "image_tensor_chw", 1, 1)
vis.save()
vis = Visualizer(save_path=f"{save_dir}/{i:06d}_2_tgt_img.png")
vis.draw(imgs[i + count // 2], "image_tensor_chw", 1, 1)
vis.save()
vis = Visualizer(save_path=f"{save_dir}/{i:06d}_3_src_feat.png")
vis.draw(features[i], "embedding_tensor_chw", 1, 1)
vis.save()
vis = Visualizer(save_path=f"{save_dir}/{i:06d}_4_tgt_feat.png")
vis.draw(features[i + count // 2], "embedding_tensor_chw", 1, 1)
vis.save()
vis = Visualizer(save_path=f"{save_dir}/{i:06d}_5_src_parse.png")
vis.draw(labels_spatial[i], "cvimage", 1, 1)
vis.save()
vis = Visualizer(save_path=f"{save_dir}/{i:06d}_6_src_parse.png")
vis.draw(labels_spatial[i + count // 2], "cvimage", 1, 1)
vis.save()
if __name__ == "__main__":
fire.Fire(main)