-
Notifications
You must be signed in to change notification settings - Fork 15
/
Copy pathinference.py
182 lines (137 loc) · 6.84 KB
/
inference.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
169
170
171
172
173
174
175
176
177
178
179
180
181
182
from model.AlignModule.generator import FaceGenerator
from model.BlendModule.generator import Generator as Decoder
from model.AlignModule.config import Params as AlignParams
from model.BlendModule.config import Params as BlendParams
from model.third.faceParsing.model import BiSeNet
import torchvision.transforms.functional as TF
import torch.nn.functional as F
import torch
import cv2
import numpy as np
import pdb
from process.process_func import Process
from process.process_utils import *
import os
import onnxruntime as ort
from utils.utils import color_transfer2
class Infer(Process):
def __init__(self,align_path,blend_path,parsing_path,params_path,bfm_folder):
Process.__init__(self,params_path,bfm_folder)
align_params = AlignParams()
blend_params = BlendParams()
self.device = 'cpu'
if torch.cuda.is_available():
self.device = 'cuda'
self.parsing = BiSeNet(n_classes=19).to(self.device)
self.netG = FaceGenerator(align_params).to(self.device)
self.decoder = Decoder(blend_params).to(self.device)
self.loadModel(align_path,blend_path,parsing_path)
self.eval_model(self.netG,self.decoder,self.parsing)
self.ort_session_sr = ort.InferenceSession('./pretrained_models/sr_cf.onnx', providers=['CPUExecutionProvider'])
def run(self,src_img_path_list,tgt_img_path_list,save_base,crop_align=False,cat=False):
os.makedirs(save_base,exist_ok=True)
i = 0
for src_img_path,tgt_img_path in zip(src_img_path_list,tgt_img_path_list):
gen = self.run_single(src_img_path,tgt_img_path,crop_align=crop_align,cat=cat)
img_name = os.path.splitext(os.path.basename(src_img_path))[0]+'-' + \
os.path.splitext(os.path.basename(tgt_img_path))[0]+'.png'
cv2.imwrite(os.path.join(save_base,img_name),gen)
print('\rhave done %04d'%i,end='',flush=True)
i += 1
print()
def run_single(self,src_img_path,tgt_img_path,crop_align=False,cat=False):
tgt_img = cv2.imread(tgt_img_path)
tgt_align = tgt_img.copy()
tgt_align,info = self.preprocess_align(tgt_img)
if tgt_align is None:
return None
src_img = cv2.imread(src_img_path)
src_align = src_img
if crop_align:
src_align,_ = self.preprocess_align(src_img,top_scale=0.55)
src_inp = self.preprocess(src_align)
tgt_inp = self.preprocess(tgt_align)
tgt_params = self.get_params(cv2.resize(tgt_align,(256,256)),
info['rotated_lmk']/2.0).unsqueeze(0)
gen = self.forward(src_inp,tgt_inp,tgt_params)
gen = self.postprocess(gen[0])
gen = self.run_sr(gen)
mask = self.mask
final = gen
# gen = color_transfer2(tgt_align,gen)
RotateMatrix = info['im'][:2]
mask = info['mask'][...,0]
rotate_gen = cv2.warpAffine(gen, RotateMatrix, (tgt_img.shape[1], tgt_img.shape[0]))
mask = cv2.warpAffine(mask, RotateMatrix, (tgt_img.shape[1], tgt_img.shape[0])) * 1.0
# ori_mask = mask.copy()
kernel2 = cv2.getStructuringElement(cv2.MORPH_RECT,(17, 17))
# mask = cv2.dilate(mask*1.0,kernel2)
mask = cv2.erode(mask*1.0,kernel2)
# mask = cv2.GaussianBlur(mask*255.0, (21, 21), 0) / 255.0
mask = cv2.blur(mask*1.0, (15, 15), 0) / 255.0
mask = np.clip(mask,0,1.0)[:,:,np.newaxis]
# pdb.set_trace()
final = rotate_gen * mask + tgt_img * (1-mask)
if cat:
final = np.concatenate([tgt_img,final],1)
final[-256:,:256] = cv2.resize(src_align,(256,256))
return final
def forward(self,xs,xt,params):
with torch.no_grad():
# xg = self.netG(F.adaptive_avg_pool2d(xs,256),
# F.adaptive_avg_pool2d(xt,256),
# params)['fake_image']
xg = F.adaptive_avg_pool2d(self.netG(F.adaptive_avg_pool2d(xs,256),
F.adaptive_avg_pool2d(xt,256),
params)['fake_image'],512)
M_a = self.parsing(self.preprocess_parsing(xg))
M_t = self.parsing(self.preprocess_parsing(xt))
M_a = self.postprocess_parsing(M_a)
M_t = self.postprocess_parsing(M_t)
# xg[M_a.repeat(1,3,1,1)==0] = -0.5
# xg[M_a.repeat(1,3,1,1)==16] = 0.6
xg_gray = TF.rgb_to_grayscale(xg,num_output_channels=1)
fake = self.decoder(xg,xg_gray,xt,M_a,M_t,xt,train=False)
gen_mask = self.parsing(self.preprocess_parsing(fake))
gen_mask = self.postprocess_parsing(gen_mask)
gen_mask = gen_mask[0][0].cpu().numpy()
mask_t = M_t[0][0].cpu().numpy()
mask = np.zeros_like(gen_mask)
for i in [1,2,3,4,5,6,7,8,9,10,11,12,13,17,18]:
mask[gen_mask==i] = 1.0
mask[mask_t==i] = 1.0
self.mask = mask
return fake
def run_sr(self,input_np):
input_np = cv2.cvtColor(input_np, cv2.COLOR_BGR2RGB)
# prepare data
input_np = input_np.transpose((2,0,1))
input_np = np.array(input_np[np.newaxis, :])
outputs_onnx = self.ort_session_sr.run(None, {'input_image':input_np.astype(np.uint8)})
out_put_onnx = outputs_onnx[0]
outimg = out_put_onnx[0,...].transpose(1,2,0)
outimg = cv2.cvtColor(outimg, cv2.COLOR_BGR2RGB)
return outimg
def loadModel(self,align_path,blend_path,parsing_path):
ckpt = torch.load(align_path, map_location=lambda storage, loc: storage)
# self.netG.load_state_dict(ckpt['G'])
self.netG.load_state_dict(ckpt['net_G_ema'])
ckpt = torch.load(blend_path, map_location=lambda storage, loc: storage)
self.decoder.load_state_dict(ckpt['G'],strict=False)
self.parsing.load_state_dict(torch.load(parsing_path))
def eval_model(self,*args):
for arg in args:
arg.eval()
if __name__ == "__main__":
model = Infer(
# 'checkpoint/Aligner/058-00008100.pth',
'pretrained_models/epoch_00190_iteration_000400000_checkpoint.pt',
'pretrained_models/Blender-401-00012900.pth',
'pretrained_models/parsing.pth',
'pretrained_models/epoch_20.pth',
'pretrained_models/BFM')
# find_path = lambda x: [os.path.join(x,f) for f in os.listdir(x)]
# img_paths = find_path('../HeadSwap/test_img')[::-1]
src_paths = ['./assets/5.jpg']
tgt_paths = ['assets/fe54875c-2cf0-4147-b08a-80552a9f46be.jpg']
model.run(src_paths,tgt_paths,save_base='res-1125',crop_align=True,cat=True)