/
test.py
142 lines (115 loc) · 5.49 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
"""
Copyright (c) 2022 Samsung Electronics Co., Ltd.
Licensed under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) License, (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at https://creativecommons.org/licenses/by-nc/4.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an
"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and limitations under the License.
For conditions of distribution and use, see the accompanying LICENSE.md file.
"""
import os
import cv2
import copy
import argparse
import numpy as np
import torch
import torch.nn.functional as F
import torch.optim as optim
from model import UNet
from dataset_raw import DatasetRAWTest
from util import calc_psnr, get_kmap_from_prob
from pytorch_msssim import ssim
parser = argparse.ArgumentParser(description='Full fixed samples')
parser.add_argument(
'--data-dir', default='/datadir', type=str, help='folder of training and validation images')
parser.add_argument(
'--file-type', default='png', type=str, help='image file type (png or tif)')
parser.add_argument(
'--checkpoint-dir', default='/checkpointdir', type=str, help='folder of checkpoint')
parser.add_argument(
'--num-iters', type=int, default=10, help='number of iterations')
parser.add_argument(
'--patch-size', type=int, default=128, help='patch size')
parser.add_argument(
'--init-features', type=int, default=32, help='init_features of UNet')
parser.add_argument(
'--k', type=float, default=1.5625, help='percentage of samples to pick')
parser.add_argument(
'--batch-size', type=int, default=1, help='batch size (DO NOT CHANGE)')
parser.add_argument(
'--lr', type=float, default=0.0001, help='learning rate')
parser.add_argument(
'--extra', type=str, default='', help='extra identifier for save folder name')
args = parser.parse_args()
grid_size = args.patch_size ** 2 * args.k / 100
if np.sqrt(grid_size) != int(np.sqrt(grid_size)):
print('Warning: superpixel grid seeds may not match the percentage of samples.')
grid_size = args.patch_size // int(np.sqrt(grid_size))
savefoldername=('k' + str(args.k)
+ '_lr' + str(args.lr)
+ '_i' + str(args.num_iters)
+ '_b' + str(args.batch_size)
+ '_ft' + str(args.init_features)
+ args.extra
)
root = os.path.join('./outputs/', savefoldername)
if not os.path.exists(root):
os.makedirs(root)
image_datasets = {x: DatasetRAWTest(os.path.join(args.data_dir,x), ftype=args.file_type)
for x in ['test']}
dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=args.batch_size,
shuffle=False, num_workers=0)
for x in ['test']}
dataset_sizes = {x: len(image_datasets[x]) for x in ['test']}
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
sampler = UNet(in_channels=6, out_channels=9, init_features=args.init_features, sigmoid=False)
sampler.load_state_dict(torch.load(os.path.join(args.checkpoint_dir, 'best_sampler.pt')))
sampler = sampler.to(device)
sampler.eval()
reconstructor = UNet(7, out_channels=3, init_features=args.init_features)
reconstructor.load_state_dict(torch.load(os.path.join(args.checkpoint_dir, 'best_reconstructor.pt')))
reconstructor = reconstructor.to(device)
reconstructor.eval()
avg_psnr = 0
avg_ssim = 0
for i, (inputs, targets) in enumerate(dataloaders['test']):
# forward: sampler
outputs = sampler(torch.cat((inputs, targets), 1))
prob = F.softmax(outputs, 1)
# sampling process
kmap = get_kmap_from_prob(prob, grid_size)
inputs2 = torch.cat([inputs, kmap * targets, kmap], dim=1)
online_sampler = copy.deepcopy(reconstructor)
online_sampler.eval()
optimizer = optim.Adam(online_sampler.parameters(), lr=args.lr)
for _ in range(args.num_iters):
optimizer.zero_grad()
outputs = online_sampler(inputs2.detach())
loss = (kmap.detach() * (outputs - targets).abs()).mean()
loss.backward()
optimizer.step()
with torch.no_grad():
outputs = online_sampler(inputs2)
# evaluation metrics
psnrout = calc_psnr(outputs, targets)
ssimout = ssim((outputs * 65535).floor(), (targets * 65535).floor(), data_range=65535, size_average=True)
avg_psnr += psnrout.item()
avg_ssim += ssimout.item()
# save images
# this is only for visualization
inputs = inputs.squeeze().permute(1, 2, 0).cpu().detach().numpy()
inputs = cv2.cvtColor(inputs, cv2.COLOR_RGB2BGR)
targets = targets.squeeze().permute(1, 2, 0).cpu().detach().numpy()
targets = cv2.cvtColor(targets, cv2.COLOR_RGB2BGR)
outputs = outputs.squeeze().permute(1, 2, 0).cpu().detach().numpy()
outputs = cv2.cvtColor(outputs, cv2.COLOR_RGB2BGR)
kmap = kmap.squeeze().cpu().detach().numpy()
cv2.imwrite(os.path.join(root, '{:07d}_in.png'.format(i)), (inputs * 255).astype(np.uint8))
cv2.imwrite(os.path.join(root, '{:07d}_gt.png'.format(i)), (targets ** (1 / 2.2) * 255).astype(np.uint8))
cv2.imwrite(os.path.join(root, '{:07d}_out.png'.format(i)), (outputs ** (1 / 2.2) * 255).astype(np.uint8))
cv2.imwrite(os.path.join(root, '{:07d}_kmap.png'.format(i)), ((1 - kmap) * 255).astype(np.uint8))
avg_psnr /= len(image_datasets['test'])
avg_ssim /= len(image_datasets['test'])
print('PSNR: {:4f}'.format(avg_psnr))
print('SSIM: {:4f}'.format(avg_ssim))