forked from e4s2024/E4S2024
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtest.py
executable file
·171 lines (131 loc) · 6.81 KB
/
test.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
"""
This file runs the main training/val loop
"""
from torch.utils.data import DataLoader
import torchvision.transforms as transforms
from datasets.dataset import CelebAHQDataset, get_transforms, TO_TENSOR, NORMALIZE, MASK_CONVERT_TF, MASK_CONVERT_TF_DETAILED
from models.networks import Net, Net2, Net3, MultiScaleNet
from models.encoder_with_optim import EncoderPlusOptimNet
from options.test_options import TestOptions
import os
import json
import sys
import pprint
import torch
from utils import torch_utils
from tqdm import tqdm
import numpy as np
from PIL import Image
sys.path.append(".")
sys.path.append("..")
class Tester:
def __init__(self,opts):
self.opts = opts
self.test_ds = CelebAHQDataset(dataset_root=self.opts.dataset_root, mode="test",
img_transform=transforms.Compose(
[TO_TENSOR, NORMALIZE]),
label_transform=transforms.Compose(
[ MASK_CONVERT_TF_DETAILED,TO_TENSOR]), # MASK_CONVERT_TF,
fraction=self.opts.ds_frac)
print(f"Number of test samples: {len(self.test_ds)}")
self.test_dataloader = DataLoader(self.test_ds, batch_size=self.opts.test_batch_size,
shuffle=False, num_workers=int(self.opts.test_workers), drop_last=False)
assert self.opts.checkpoint_path is not None, "please specify the pre-trained weights!"
self.net = Net3(self.opts).eval().to(self.opts.device)
ckpt_dict=torch.load(self.opts.checkpoint_path)
self.net.latent_avg = ckpt_dict['latent_avg'].to(self.opts.device) if self.opts.start_from_latent_avg else None
self.net.load_state_dict(torch_utils.remove_module_prefix(ckpt_dict["state_dict"],prefix="module."))
print("Load pre-trained weights.")
# @torch.no_grad()
# def test(self,num_imgs=20, save_dir=""):
# cnt = 0
# for batch_idx, batch in enumerate(self.test_dataloader):
# (img1, img2), (mask1, mask2), (mask1_vis, mask1_vis) = batch
# img1, img2 = img1.to(self.opts.device).float(), img2.to(
# self.opts.device).float()
# mask1 = (mask1*255).long().to(self.opts.device)
# mask2 = (mask2*255).long().to(self.opts.device)
# # [bs,1,H,W]的mask 转成one-hot格式，即[bs,#seg_cls,H,W]
# onehot1 = torch_utils.labelMap2OneHot(
# mask1, num_cls=self.opts.num_seg_cls)
# onehot2 = torch_utils.labelMap2OneHot(
# mask2, num_cls=self.opts.num_seg_cls)
# swap = False
# recon1, latent1 = self.net(img1, onehot1, return_latents=True)
# recon2, latent2 = self.net(img2, onehot2, return_latents=True)
# imgs_1 = self.parse_images(onehot1, img1, recon1)
# imgs_2 = self.parse_images(onehot2, img2, recon2)
# self.save_imgs(imgs_1, save_name="%04d" % (cnt))
# self.save_imgs(imgs_2, save_name="%04d" % (cnt+1))
# cnt += len(img1)*2
# if cnt>num_imgs:
# break
# 补偿网络测试
@torch.no_grad()
def compensate_test(self,num_imgs=20, save_dir=""):
cnt = 0
for batch_idx, batch in enumerate(tqdm(self.test_dataloader)):
img, recon_img, optimed_style_code, mask, mask_vis = batch
sample_name = os.path.basename(self.test_ds.imgs[batch_idx])[:-4]
img = img.to(self.opts.device).float()
recon_img = recon_img.to(self.opts.device).float()
optimed_style_code = optimed_style_code.to(self.opts.device).float()
mask = (mask*255).long().to(self.opts.device)
# [bs,1,H,W]的mask 转成one-hot格式，即[bs,#seg_cls,H,W]
onehot = torch_utils.labelMap2OneHot(mask, num_cls=self.opts.num_seg_cls)
diffMap = img - recon_img
res_feats, base_codes, compensated_codes = self.net.get_style(img, onehot, diff_img=diffMap)
# 1. 不用补偿style code 的重建结果
recon_1 = self.net.gen_img(res_feats, base_codes, onehot)
# 2. 用了补偿style code 的重建结果
recon_2 = self.net.gen_img(res_feats, base_codes + compensated_codes, onehot)
imgs = {
'input_face': torch_utils.tensor2im(img[0]),
'input_mask': torch_utils.tensor2map(onehot[0]),
'recon_face_wo_compensation': torch_utils.tensor2im(recon_1[0]),
'recon_face_w_compensation': torch_utils.tensor2im(recon_2[0]),
}
self.save_imgs(imgs, save_name=sample_name)
cnt += len(img)
if cnt>num_imgs:
break
@torch.no_grad()
def test(self,num_imgs=20, save_dir="",begin_idx=0):
cnt = 0
for batch_idx, batch in enumerate(tqdm(self.test_dataloader)):
if batch_idx < begin_idx:
continue
img, mask, mask_vis = batch
sample_name = os.path.basename(self.test_ds.imgs[batch_idx])[:-4]
img = img.to(self.opts.device).float()
mask = (mask*255).long().to(self.opts.device)
# [bs,1,H,W]的mask 转成one-hot格式，即[bs,#seg_cls,H,W]
onehot = torch_utils.labelMap2OneHot(mask, num_cls=self.opts.num_seg_cls)
recon, structure_codes_GT, latent = self.net(img, onehot, return_latents=True)
imgs = self.parse_images(onehot, img, recon)
self.save_imgs(imgs, save_name=sample_name)
cnt += len(img)
if cnt>num_imgs:
break
def save_imgs(self, img_data, save_name):
# # gt 和 recon 放在一张图·
# arr = np.hstack((
# np.array(img_data["input_face"]),
# np.array(img_data["recon_face"])
# ))
# Image.fromarray(arr).save(os.path.join(self.opts.save_dir, "%s.png" % save_name))
for k, v in img_data.items():
if k=="recon_face":
v.save(os.path.join(self.opts.save_dir, "%s_%s.png" % (save_name, k)))
def parse_images(self, mask, img, recon):
cur_im_data = {
'input_face': torch_utils.tensor2im(img[0]),
'input_mask': torch_utils.tensor2map(mask[0]),
'recon_face': torch_utils.tensor2im(recon[0]),
}
return cur_im_data
if __name__ == '__main__':
opts = TestOptions().parse()
os.makedirs(opts.save_dir, exist_ok=True)
tester = Tester(opts)
tester.test(num_imgs=2000,begin_idx=0)