-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathtrain_distributed_launch.py
196 lines (130 loc) · 6.45 KB
/
train_distributed_launch.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
import datetime
import os
import math
import sys
import csv
import random
import numpy as np
import torch
import torch.multiprocessing as mp
from hyper_para import HyperParameters
import torch.distributed as distributed
from torch import nn
from util.logger import TensorboardLogger
from model.model import PropagationModel
from dataset.RefCOCO_dataset import RefCOCOTrainDataset #
from dataset.RefCOCO_dataset import RefCOCOTestDataset
from dataset.YoutubeVOS_dataset import YoutubevosTrainDataset
from dataset.YoutubeVOS_dataset import YoutubevosTestDataset
from dataset.A2D_dataset import A2DTrainDataset, A2DTestDataset
from a2d_test import a2d_evaluate
from dataset.DAVIS_dataset import DAVISTrainDataset
from dataset.DAVIS_dataset import DAVISTestDataset
from refcoco_test import refcoco_evaluate
from davis_test import davis_evaluate
from model.network import PropagationNetwork
from util.train_utils import renew_loader, evaluate_during_training
para = HyperParameters()
para.parse()
num_gpus = torch.cuda.device_count()
distributed.init_process_group(backend="nccl")
print('CUDA Device count: ', torch.cuda.device_count())
if para['benchmark']:
torch.backends.cudnn.benchmark = True
local_rank = torch.distributed.get_rank()
world_size = torch.distributed.get_world_size()
torch.cuda.set_device(local_rank)
print('I am rank %d in this world of size %d!' % (local_rank, world_size))
if para['id'].lower() != 'null':
print('I will take the role of logging!')
long_id = para['id']
else:
long_id = None
logger = TensorboardLogger(para['id'], long_id)
logger.log_string('hyperpara', str(para))
save_path = os.path.join(para["save_mdoel_dir"], para["folder_name"], long_id, long_id) if long_id is not None else None
output_dir = os.path.join(para["save_mdoel_dir"], para["folder_name"], long_id) if long_id is not None else None
if local_rank == 0:
model = PropagationModel(para, logger=logger, save_path=save_path, local_rank=local_rank, world_size=world_size, distributed=True)
else:
model = PropagationModel(para, local_rank=local_rank, world_size=world_size, distributed=True).train()
model_test = PropagationNetwork(para).cuda()
total_iter = 0
##########################data load########################
skip_values = [10, 15, 20, 25, 5]
if para['stage'] == 0: # RefCOCO
dataset_root = os.path.expanduser(para['refcoco_root'])
train_dataset, train_sampler, train_loader = renew_loader(para, dataset_root, local_rank=local_rank)
test_dataset = RefCOCOTestDataset(para, dataset_root, split="val")
print('RefCOCO dataset size: ', len(train_dataset))
elif para['stage'] == 1: # YoutubeVOS
increase_skip_fraction = [0.1, 0.2, 0.3, 0.4, 0.9, 1.0]
max_skip = 5
a2d_dataset_root = para["a2d_root"]
a2d_test_dataset = A2DTestDataset(para, a2d_dataset_root)
train_dataset, train_sampler, train_loader = renew_loader(para, max_skip=5, local_rank=local_rank)
test_datasets = [a2d_test_dataset]
else:
RuntimeError("Wrong stage")
total_epoch = math.ceil(para['iterations']/len(train_loader))
current_epoch = total_iter // len(train_loader)
print('Number of training epochs (the last epoch might not complete): ', total_epoch)
if para['stage'] == 0:
epoch_start_eval = total_epoch - 10
elif para['stage'] == 1:
epoch_start_eval = total_epoch - 1
elif para['stage'] == 2:
epoch_start_eval = 0
print('The epoch that we start to eval: ', epoch_start_eval)
if para['stage'] != 0:
increase_skip_epoch = [round(total_epoch * f) for f in increase_skip_fraction]
print('The skip value will increase approximately at the following epochs: ', increase_skip_epoch[:-1])
try:
IoU = 0
for e in range(current_epoch, total_epoch):
print('Epoch %d/%d' % (e, total_epoch))
# reset the skip value
if para['stage'] != 0 and e != total_epoch and e >= increase_skip_epoch[0]:
while e >= increase_skip_epoch[0]:
cur_skip = skip_values[0]
skip_values = skip_values[1:]
increase_skip_epoch = increase_skip_epoch[1:]
print('Increasing skip to: ', cur_skip)
if para['stage'] != 0:
train_dataset, train_sampler, train_loader = renew_loader(para=para, max_skip=cur_skip, local_rank=local_rank)
train_sampler.set_epoch(e)
model.train()
for data in train_loader:
model.do_pass(data, total_iter)
total_iter += 1
if total_iter >= para['iterations']:
break
if local_rank == 0:
if total_iter >= 16000:
model.save(total_iter)
model_test.load_state_dict(model.model.module.state_dict())
test_samplers = [torch.utils.data.SequentialSampler(test_dataset) for test_dataset in test_datasets]
data_loader_tests = [
torch.utils.data.DataLoader(test_datasets[k], batch_size=1, sampler=test_samplers[k], num_workers=8) for
k in range(len(test_datasets))]
mean_iou, overall_iou, precision5, precision6, precision7, precision8, precision9, precision_mAP, total_time = evaluate_during_training(
args=para, test_datasets=test_datasets, model_test=model_test, data_loader_tests=data_loader_tests,
epoch=e,
long_id=long_id, epoch_start_eval=epoch_start_eval, iterations=total_iter, output_dir=output_dir)
metrics = [total_iter, precision5, precision6, precision7, precision8, precision9, precision_mAP, overall_iou, mean_iou]
columns_name = ["iteration", "precision5", "precision6", "precision7", "precision8", "precision9",
"precision_mAP", "overall_iou", "mean_iou"]
if not os.path.exists("log/" + long_id + "/eval_{}.csv".format(long_id)):
with open("log/" + long_id + "/eval_{}.csv".format(long_id), mode='w', newline='', encoding='utf8') as cfa:
pf = csv.writer(cfa)
pf.writerow(columns_name)
pf.writerow(metrics)
else:
with open("log/" + long_id + "/eval_{}.csv".format(long_id), mode='a', newline='', encoding='utf8') as cfa:
pf = csv.writer(cfa)
pf.writerow(metrics)
finally:
if not para['debug'] and model.logger is not None and total_iter>5000:
model.save(total_iter)
# Clean up
distributed.destroy_process_group()