-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
209 lines (172 loc) · 8.28 KB
/
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
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
# torchrun --nproc_per_node=2 --master_port=29501 home/jthe/blur_detection/blur_detector/main_ddp.py
import os
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import time
import sys
import cv2
import random
import argparse
from PIL import Image
from torch.utils.tensorboard import SummaryWriter
from tqdm import tqdm
from torch.utils.data.distributed import DistributedSampler
import torch.distributed as dist
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = True
# set random seed
torch.manual_seed(39)
torch.cuda.manual_seed(39)
random.seed(39)
np.random.seed(39)
from torch.utils.data import DataLoader
from dataset.dataloader import BlurMagDataset
from utils.logger import Logger
from model.bme_model import MyNet_Res50_multiscale
from train.optimizer import Optimizer
class Trainer():
def __init__(self, args) -> None:
self.args = args
if not self.args.test_only:
self.model_setting()
self.dataset_setting()
self.weight_path = self.args.weight_path
os.makedirs(self.weight_path, exist_ok=True)
if dist.get_rank() == 0:
self.logger = Logger(self.args.logger_path)
self.l1_loss = torch.nn.L1Loss()
self.opt = Optimizer(self.model, self.args)
self.min_loss = float('inf')
def train(self):
for epoch in range(self.args.epochs):
self.model.train()
train_loss = self.train_one_epoch(epoch)
test_loss = self.test(epoch)
if dist.get_rank() == 0:
self.logger.register(epoch, train_loss, test_loss)
if (test_loss < self.min_loss):
self.min_loss = test_loss
torch.save(self.model.module.state_dict(), os.path.join(self.weight_path, "best_net.pth"))
self.logger.save_best(epoch)
if (epoch+1) % 10 == 0:
torch.save(self.model.module.state_dict(), os.path.join(self.weight_path, f"epoch_{epoch+1}.pth"))
def train_one_epoch(self, epoch):
total_loss = 0
num_samples = 0
self.train_sampler.set_epoch(epoch)
pbar = tqdm(self.train_dl, total=len(self.train_dl))
pbar.set_description(f'Epoch [{epoch}/{self.args.epochs}] training')
for i,data in enumerate(pbar):
blur_img, blur_mag = data
blur_img, blur_mag = blur_img.to(self.args.device),blur_mag.to(self.args.device)
pred_mag = self.model(blur_img)
loss = self.l1_loss(pred_mag, blur_mag) + self.sill_loss(pred_mag, blur_mag)
loss.backward()
self.opt.step()
self.opt.zero_grad()
total_loss += loss.detach().item()
num_samples += 1
avg_loss = total_loss / num_samples
pbar.set_description(f"Epoch {epoch+1} Loss: {avg_loss:.4f} Lr: {self.opt.get_lr()}")
pbar.update(1)
pbar.close()
self.opt.lr_schedule()
return avg_loss
def test(self, epoch):
self.model.eval()
with torch.no_grad():
total_loss = 0
num_samples = 0
self.test_sampler.set_epoch(epoch)
pbar = tqdm(self.test_dl, total=len(self.test_dl))
pbar.set_description(f'Epoch [{epoch}/{self.args.epochs}] testing')
for i,data in enumerate(pbar):
blur_img, blur_mag = data
blur_img, blur_mag = blur_img.to(self.args.device),blur_mag.to(self.args.device)
pred_mag = self.model(blur_img)
pred_mag = pred_mag.squeeze(1)
loss = self.l1_loss(pred_mag, blur_mag)
total_loss += loss.detach().item()
num_samples += 1
avg_loss = total_loss / num_samples
pbar.set_description(f"Epoch {epoch+1} Loss: {avg_loss:.4f}")
pbar.update(1)
pbar.close()
return avg_loss
def inference(self):
# load model
checkpoint_path = os.path.join(self.args.weight_path, self.args.model_name)
checkpoint = torch.load(checkpoint_path, map_location=lambda storage, loc: storage.cuda())
self.model = MyNet_Res50_multiscale().cuda()
self.model.load_state_dict(checkpoint)
print("Loading Model Done")
# load dataset
infer_dataset = BlurMagDataset(dataset_root=self.args.infer_dataset_path,train=False)
infer_dl = DataLoader(infer_dataset, batch_size=1, shuffle=False, pin_memory=True)
print("Loading Dataset Done")
output_folder = self.args.infer_output_path
os.makedirs(output_folder, exist_ok=True)
print("Starting Evaluation")
self.model.eval()
with torch.no_grad():
for i,data in enumerate(infer_dl):
print("Complete: ", i)
blur_img, video_name, file_name = data
blur_img = blur_img.cuda()
output = self.model(blur_img)
# output = output.clamp(-0.5, 0.5)
output = output.clamp(0 ,1)
output = output[0].to('cpu').detach().numpy().squeeze()
# output = ((output+0.5) * 205)
output = ((output) * 205)
output = output/np.max(output)
output = np.uint8(255-(output*255))
output_video_folder = os.path.join(output_folder, video_name[0])
os.makedirs(output_video_folder, exist_ok=True)
output_path = os.path.join(output_video_folder,file_name[0])
cv2.imwrite(output_path, output)
# output.save(output_path)
def model_setting(self):
if self.args.local_rank != -1:
torch.cuda.set_device(self.args.local_rank)
device = torch.device("cuda", self.args.local_rank)
dist.init_process_group(backend="nccl", init_method='env://')
self.args.device = device
print("device:", self.args.device)
num_gpus = torch.cuda.device_count()
print("# of gpus",num_gpus)
self.model = MyNet_Res50_multiscale()
self.model.to(self.args.device)
self.model = nn.parallel.DistributedDataParallel(self.model, device_ids=[self.args.local_rank], output_device=self.args.local_rank)
def dataset_setting(self):
dataset_train = BlurMagDataset(dataset_root=self.args.training_dataset_path,train=True)
self.train_sampler = DistributedSampler(dataset_train)
self.train_dl = DataLoader(dataset_train, sampler=self.train_sampler, batch_size=self.args.batch_size,pin_memory=True,
drop_last=True,num_workers=4)
dataset_test = BlurMagDataset(dataset_root=self.args.testing_dataset_path,train=True)
self.test_sampler = DistributedSampler(dataset_test)
self.test_dl = DataLoader(dataset_test, sampler=self.test_sampler, batch_size=self.args.batch_size,pin_memory=True,
drop_last=True,num_workers=4)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--local_rank", default=os.getenv('LOCAL_RANK', -1), type=int)
parser.add_argument("--weight_path", default="home/jthe/BME/BME/weights/", type=str)
parser.add_argument("--logger_path", default="home/jthe/BME/BME/log/myunet_v3/training_loss.txt", type=str)
parser.add_argument("--training_dataset_path", default="disk2/jthe/datasets/GOPRO_blur_magnitude/train", type=str)
parser.add_argument("--testing_dataset_path", default="disk2/jthe/datasets/GOPRO_blur_magnitude/test", type=str)
parser.add_argument("--infer_dataset_path", default="disk2/jthe/datasets/GOPRO_blur_magnitude/test/frame11", type=str)
parser.add_argument("--infer_output_path", default="home/jthe/BME/BME/output/", type=str)
parser.add_argument("--epochs", default=500, type=int)
parser.add_argument("--init_lr", default=1e-3, type=float)
parser.add_argument("--final_lr", default=1e-5, type=float)
parser.add_argument("--batch_size", default=8, type=int)
parser.add_argument('--test_only', action='store_true')
parser.add_argument("--model_name", default="best_net.pth", type=str)
args = parser.parse_args()
trainer = Trainer(args)
if args.test_only:
trainer.inference()
else:
trainer.train()