-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
116 lines (103 loc) · 5.04 KB
/
train.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
import os
import cv2
import math
import time
import torch
import torch.distributed as dist
import numpy as np
import random
import argparse
from model.VFSS_train_b import Model
from model.dataset import *
from torch.utils.data import DataLoader, Dataset
from torch.utils.tensorboard import SummaryWriter
from torch.utils.data.distributed import DistributedSampler
device = torch.device("cuda")
log_path = 'log'
def get_learning_rate(step):
if step < 2000:
mul = step / 2000.
return 3e-4 * mul
else:
mul = np.cos((step - 2000) / (args.epoch * args.step_per_epoch - 2000.) * math.pi) * 0.5 + 0.5
return (3e-4 - 3e-6) * mul + 3e-6
def flow2rgb(flow_map_np):
h, w, _ = flow_map_np.shape
rgb_map = np.ones((h, w, 3)).astype(np.float32)
normalized_flow_map = flow_map_np / (np.abs(flow_map_np).max())
rgb_map[:, :, 0] += normalized_flow_map[:, :, 0]
rgb_map[:, :, 1] -= 0.5 * (normalized_flow_map[:, :, 0] + normalized_flow_map[:, :, 1])
rgb_map[:, :, 2] += normalized_flow_map[:, :, 1]
return rgb_map.clip(0, 1)
def train(model, local_rank):
if local_rank == 0:
writer = SummaryWriter('log/train', purge_step=args.start_step)
else:
writer = None
step = args.start_step
nr_eval = 0
dataset = VimeoDataset('train')
train_data = DataLoader(dataset, batch_size=args.batch_size, shuffle=True, num_workers=6, pin_memory=True,
drop_last=True)
args.step_per_epoch = train_data.__len__()
time_stamp = time.time()
for epoch in range(args.start_epoch, args.epoch):
for i, data in enumerate(train_data):
data_time_interval = time.time() - time_stamp
time_stamp = time.time()
data_gpu, timestep = data
data_gpu = data_gpu.to(device, non_blocking=True) / 255.
timestep = timestep.to(device, non_blocking=True)
imgs = data_gpu[:, :12]
gt = data_gpu[:, 12:15]
learning_rate = get_learning_rate(step)
pred, flow, metric0, metric1, loss_l1, loss_lpips = model.update(imgs, gt, learning_rate, True, timestep,
step, args.step_per_epoch)
train_time_interval = time.time() - time_stamp
time_stamp = time.time()
if step % 20 == 1 and local_rank == 0:
writer.add_scalar('learning_rate', learning_rate, step)
writer.add_scalar('loss/l1', loss_l1, step)
writer.add_scalar('loss/lpips', loss_lpips, step)
if step % 100 == 1 and local_rank == 0:
gt = (gt.permute(0, 2, 3, 1).detach().cpu().numpy() * 255).astype('uint8')
metric = (torch.cat((metric0, metric1), 3).permute(0, 2, 3, 1).detach().cpu().numpy() * 255).astype(
'uint8')
pred = (pred.permute(0, 2, 3, 1).detach().cpu().numpy() * 255).astype('uint8')
flow = flow.permute(0, 2, 3, 1).detach().cpu().numpy()
for i in range(5):
imgs = np.concatenate((pred[i], gt[i]), 1)
writer.add_image(str(i) + '/img', imgs, step, dataformats='HWC')
writer.add_image(str(i) + '/flow', flow2rgb(flow[i]), step, dataformats='HWC')
writer.add_image(str(i) + '/metric', metric[i], step, dataformats='HWC')
writer.flush()
if step % 1000 == 1 and local_rank == 0:
model.save_model(log_path, epoch=epoch, step=i, rank=local_rank)
# time.sleep(1)
if local_rank == 0:
print('epoch:{} {}/{} time:{:.2f}+{:.2f} loss_l1:{:.4e} loss_lpips:{:.4e}'.format(epoch, i, args.step_per_epoch,
data_time_interval, train_time_interval,
loss_l1, loss_lpips))
step += 1
nr_eval += 1
model.save_model(log_path, rank=local_rank)
print(f'save model at e{epoch}s{step}')
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--epoch', default=12, type=int)
parser.add_argument('--batch_size', default=16, type=int, help='minibatch size')
parser.add_argument('--local_rank', default=0, type=int, help='local rank')
parser.add_argument('--world_size', default=1, type=int, help='world size')
parser.add_argument('--start_epoch', default=0, type=int, help='start_epoch')
parser.add_argument('--start_step', default=0, type=int, help='start_step')
args = parser.parse_args()
torch.cuda.set_device(args.local_rank)
seed = 1234
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.benchmark = True
model = Model(mode='MOF')
model.load_model('train_log')
train(model, args.local_rank)