-
Notifications
You must be signed in to change notification settings - Fork 10
/
test_main.py
187 lines (153 loc) · 6.95 KB
/
test_main.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
183
184
185
186
187
"""
Training script for Multiple Objectives Network (MoNet)
"""
from __future__ import print_function
import argparse
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import os
import torch.optim
import torch.utils.data
import torch.utils.data.distributed
from models import HyperRes
from utils import AverageMeter
from utils.DataUtils import CommonTools
from utils.DataUtils.CommonTools import calculate_psnr, postProcessForStats, saveImage
from utils.DataUtils.TrainLoader import NoisyDataset
import numpy as np
from torchmetrics.image.lpip import LearnedPerceptualImagePatchSimilarity
parser = argparse.ArgumentParser(description='PyTorch ImageNet Testing')
parser.add_argument('--data', metavar='DIR', help='path to dataset')
parser.add_argument('--workers', default=16, type=int, metavar='N',
help='number of data loading workers (default: 4)')
parser.add_argument('--weights', required=True, type=str, metavar='PATH',
help='Path to latest checkpoint')
parser.add_argument('--meta_blocks', type=int, default=16, help='Number of Meta Blocks (default: 16)')
parser.add_argument('--valid', type=str, default='valid', help='Evaluation folder')
parser.add_argument('--device', type=str, default='cpu', help='Device to run on,[cpu,cuda..]')
parser.add_argument('--lvls', type=int, nargs='+', default=[15], help='A list of corruptions levels to evaluate')
parser.add_argument('-y', '-Y', '--gray', dest='y_channel', default=False, action='store_true',
help='Test on Grayscale')
parser.add_argument('--sample', dest='save_sample', default=False, action='store_true', help='Save sample images')
parser.add_argument('--data_type', type=str, default='n', choices=['n', 'sr', 'j'],
help='Defines the task data, de(n)oise, super-resolution(sr), de(j)peg.')
parser.add_argument('--norm_f', type=float, default=255, help='The normalization factor for the distortion levels.')
def getBounds(lvls_arr: list, n_lvl: float):
"""
A helper function to calculate the BN interpolation
@param lvls_arr: The corruption levels list
@param n_lvl: The current level to interpolate
@return: Returns the post and previous corruption levels
"""
lower = lvls_arr[0]
upper = lvls_arr[1]
for i, v in enumerate(lvls_arr[:-1]):
if n_lvl <= v:
break
lower = v
upper = lvls_arr[i + 1]
return lower, upper
def bn_interpolate(state_dict: dict, orig_lvls: list, new_lvls: list):
"""
Interpolates the Batch Normalization parameters
@param state_dict: The original model weights
@param orig_lvls: Trained corruption levels
@param new_lvls: Inference corruption levels
@return: Model weights with modified BN params
"""
new_state_dict = {}
for key in state_dict.keys():
if 'bn_' in key:
for sig in new_lvls:
if len(orig_lvls) > 1:
lower, upper = getBounds(orig_lvls, sig)
if sig > upper:
alpha = 0
beta = 1
elif sig < lower:
alpha = 1
beta = 0
else:
alpha = (upper - sig) / (upper - lower)
beta = 1 - alpha
upper_key = key[:key.find('bn_') + 3] + str(upper) + key[key.rfind('.'):]
lower_key = key[:key.find('bn_') + 3] + str(lower) + key[key.rfind('.'):]
upper_tensor = state_dict[upper_key]
lower_tensor = state_dict[lower_key]
new_value = lower_tensor * alpha + beta * upper_tensor
else:
new_value = state_dict[key]
new_key = key[:key.find('bn_') + 3] + str(sig) + key[key.rfind('.'):]
new_state_dict[new_key] = new_value
else:
new_state_dict[key] = state_dict[key]
return new_state_dict
def main():
global args
args = parser.parse_args()
# Create model
model = HyperRes(meta_blocks=args.meta_blocks, level=args.lvls, device=args.device,
gray=args.y_channel, norm_factor=args.norm_f)
# Load weights
if not os.path.isfile(args.weights):
print("=> no checkpoint found at '{}'".format(args.weights))
exit()
print("=> loading weights '{}'".format(args.weights))
checkpoint = torch.load(args.weights, map_location=args.device)
orig_sigmas = checkpoint['sigmas']
new_weights = bn_interpolate(checkpoint['state_dict'], orig_sigmas, args.lvls)
model.load_state_dict(new_weights)
model.to(args.device)
CommonTools.set_random_seed(0)
torch.backends.cudnn.benckmark = True
# Data loading code
val_loader = NoisyDataset(
os.path.join(args.data, args.valid),
cor_lvls=args.lvls,
phase='test',
interp=False,
lr_prefix=args.data_type,
)
val_loader = torch.utils.data.DataLoader(
val_loader,
batch_size=1, shuffle=False,
num_workers=1, pin_memory=True)
psnr_avg, lipips_avg = evaluate(val_loader, model)
print("Results:")
for k, v in psnr_avg.items():
print("Sigma {}:\t {:4f}".format(k, v))
print("LIPIPS:")
for k, v in lipips_avg.items():
print("Sigma {}:\t {:4f}".format(k, v))
def evaluate(val_loader, model):
print("Evaluation:")
sig_loss = [AverageMeter() for _ in range(len(args.lvls))]
lpips_loss = [AverageMeter() for _ in range(len(args.lvls))]
lpips = LearnedPerceptualImagePatchSimilarity(net_type='vgg')
# switch to evaluate mode
model.eval()
with torch.no_grad():
random_image = np.random.randint(len(val_loader)) if args.save_sample else -1
for i, data in enumerate(val_loader):
target = data['target'].to(args.device)
images = [x.to(args.device) for x in data['image']]
if i == random_image:
saveImage(output[j].detach().cpu().numpy()[0], target.detach().cpu().numpy()[0], j)
# Compute output
output = model(images)
for j in range(len(args.lvls)):
# Measure PSNR
for out_idx, out in enumerate(output[j]):
imgs = postProcessForStats([target[out_idx], out])
trg, out = imgs
psnr = calculate_psnr(out, trg, args.data_type == 'sr')
lpips_loss[j].update(
lpips(
torch.FloatTensor(np.expand_dims(out.transpose((2,0,1)),0)/255.0),
torch.FloatTensor(np.expand_dims(trg.transpose((2,0,1)),0)/255.0)
)
)
sig_loss[j].update(psnr)
return {s: t.avg for s, t in zip(args.lvls, sig_loss)}, {s: t.avg for s, t in zip(args.lvls, lpips_loss)}
if __name__ == '__main__':
main()