-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathtrain_and_eval.py
193 lines (172 loc) · 6.84 KB
/
train_and_eval.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
188
189
190
191
192
193
import functools
import math
from collections import defaultdict
from PIL import Image
import tqdm
import numpy as np
import torch
import glob
import time
from loss import ContentLoss
from metrics import psnr_fn, ssim_fn
from model import ZSSRModel, ZSSRModelWithBackbone
from data import ZSSRDataset, ZSSRSampler, VDSRDataset
from torchvision.transforms import transforms
from torch import nn
from transforms import ToTensor, RandomCrop
import torch.nn.functional as F
from torch.utils.data import DataLoader
from config import config
def train_single_img(lr_img,
model,
upsample,
data_sampler,
trans,
optimizer,
num_batches,
device="cuda"):
sr_factor = config["sr_factor"] if config is not None else 2
model.train()
avg_loss = 0
loss_type = config["loss_type"] if config is not None else "l1"
l1_loss = F.l1_loss
if loss_type == "content" or loss_type == "hybrid":
content_loss = ContentLoss()
lr_downsampled = lr_img.resize([lr_img.size[0] // sr_factor, lr_img.size[1] // sr_factor],
resample=Image.BICUBIC)
lr_upsampled = lr_downsampled.resize([lr_downsampled.size[0] * sr_factor, lr_downsampled.size[1] * sr_factor],
resample=Image.BICUBIC)
lr_upsampled = transforms.ToTensor()(lr_upsampled).unsqueeze(0).to(device)
lr_img_tensor = transforms.ToTensor()(lr_img).unsqueeze(0).to(device)
for iter, (hr, lr) in enumerate(data_sampler):
optimizer.zero_grad()
scale = hr.size[0] // lr.size[0]
lr = upsample(lr, scale)
hr, lr = trans((hr, lr))
hr, lr = hr.unsqueeze(0).to(device), lr.unsqueeze(0).to(device)
hr_pred = model(lr)
if loss_type == "content":
loss = content_loss(model(lr_upsampled), lr_img_tensor)
elif loss_type == "hybrid":
loss = config["l1_loss_coff"] * l1_loss(hr_pred, hr)
loss += config["content_loss_coff"] * content_loss(model(lr_upsampled), lr_img_tensor)
else:
loss = l1_loss(hr_pred, hr)
avg_loss += loss.item()
loss.backward()
optimizer.step()
if iter > num_batches:
print('Done training.')
avg_loss /= iter
print(f'Avg training loss is {avg_loss}')
break
def test_single_img(model,
upsample,
downsampled_img,
original_img,
sr_factor):
model.eval()
with torch.set_grad_enabled(False):
lr_upsampled = upsample(downsampled_img, sr_factor)
hr_pred = model(lr_upsampled.unsqueeze(0).cuda()).squeeze(0)
hr_pred = hr_pred.cpu().numpy()
hr_pred[np.where(hr_pred < 0)] = 0.0
hr_pred[np.where(hr_pred > 1)] = 1.0
original_img = np.array(original_img) / 255.0
original_img = np.transpose(original_img, (2, 0, 1))
metrics = {}
if "psnr" in config["metrics"]:
metrics["psnr"] = psnr_fn(hr_pred, original_img)
if "ssim" in config["metrics"]:
metrics["ssim"] = ssim_fn(hr_pred, original_img)
return metrics
def train_test_single_img(hr_img_path, lr_img_path, config):
sr_factor = config["sr_factor"]
img = Image.open(hr_img_path)
lr_img = Image.open(lr_img_path)
print(f"Starting training on {hr_img_path} with resolution factor {sr_factor}")
dataset = ZSSRDataset.from_image(lr_img, config["sr_factor"])
data_sampler = ZSSRSampler(dataset)
if config["model"] == "zssr":
model = ZSSRModel()
else:
vdsr_backbone = torch.load("./models/model_epoch_100.pth")
vdsr_backbone.to(config["device"])
model = ZSSRModelWithBackbone(vdsr_backbone, True)
model.to(config["device"])
model.train()
all_models = None
trans = transforms.Compose([
ToTensor(),
RandomCrop(config["crop_size"])
])
if config["upsample"] == "pixelshuffle":
trans = transforms.Compose([
ToTensor(),
# RandomCrop(config["crop_size"])
])
upsamples_layers = []
for _ in range(int(math.log(config["sr_factor"], 2))):
upsamples_layers += [nn.Conv2d(3, 3 * 4, kernel_size=3, padding=3 // 2, bias=True),
nn.ReLU(),
nn.PixelShuffle(2)]
upsample_model = nn.Sequential(*upsamples_layers)
upsample_model.to(config["device"])
upsample_model.train()
upsample_fn = functools.partial(pixelshuffle_upsample, upsample_model=upsample_model)
# we do some of the computation in the upsample phase
model = ZSSRModel(layers_num=6 - int(math.log(sr_factor)))
model.to(config["device"])
model.train()
all_models = nn.ModuleList([upsample_model, model]).to(config["device"])
elif config["upsample"] == "cubic":
all_models = model
upsample_fn = bicubic_upsample
optimizer = torch.optim.Adam(all_models.parameters(), lr=config["learning_rate"])
num_batches = config["number_of_iterations"]
train_single_img(lr_img,
model,
upsample_fn,
data_sampler,
trans,
optimizer,
num_batches)
model.eval()
if config["upsample"] != "cubic":
upsample_model.eval()
metrics = test_single_img(model,
upsample_fn,
lr_img,
img,
config["sr_factor"])
return metrics
def pixelshuffle_upsample(lr, scale, upsample_model):
lr = transforms.ToTensor()(lr).unsqueeze(0)
lr = lr.cuda()
lr = upsample_model(lr)
return lr.squeeze(0)
def bicubic_upsample(lr, scale):
lr = lr.resize([lr.size[0] * scale, lr.size[1] * scale],
resample=Image.BICUBIC)
return transforms.ToTensor()(lr)
def train_test_all(hr_folder_name, lr_folder_name, config):
hr_images = sorted(glob.glob(f"{hr_folder_name}/img_*.png"))
lr_images = sorted(glob.glob(f"{lr_folder_name}/img_*.png"))
all_metrics = {}
for metric_name in config["metrics"]:
all_metrics[metric_name] = []
for hr_path, lr_path in zip(hr_images, lr_images):
metrics = train_test_single_img(hr_path, lr_path, config)
print(f"Finish evaluation")
for k, v in metrics.items():
print(f"Average {k}: {v}")
all_metrics[k].append(v)
for k, metric in all_metrics.items():
avg_metric = sum(metric) / len(metric)
print(f"Average {k} on all images: {avg_metric}")
if __name__ == "__main__":
from PIL import Image
config["sr_factor"] = 2
config["upsample"] = "cubic"
config["loss_type"]="content"
train_test_all("./data/BSD100", "bicubic_kernel_sr2", config)