-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain_ofa_stereo.py
268 lines (239 loc) · 9.87 KB
/
train_ofa_stereo.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
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
# Once for All: Train One Network and Specialize it for Efficient Deployment
# Han Cai, Chuang Gan, Tianzhe Wang, Zhekai Zhang, Song Han
# International Conference on Learning Representations (ICLR), 2020.
import argparse
import numpy as np
import os
import random
import horovod.torch as hvd
import torch
from ofa.stereo_matching.elastic_nn.modules.dynamic_op import DynamicSeparableConv2d
from ofa.stereo_matching.elastic_nn.networks import OFAAANet
from ofa.stereo_matching.run_manager import DistributedStereoRunConfig
from ofa.stereo_matching.run_manager.distributed_run_manager import DistributedRunManager
from ofa.utils import download_url, MyRandomResizedCrop
from ofa.stereo_matching.elastic_nn.training.progressive_shrinking import load_models
parser = argparse.ArgumentParser()
parser.add_argument('--task', type=str, default='large', choices=[
'kernel',
'depth',
'expand',
'scale',
'large',
'final',
'kitti2012',
'kitti2015',
'kitti_mix',
'kitti_sf_mix',
])
parser.add_argument('--phase', type=int, default=1, choices=[1, 2])
parser.add_argument('--resume', action='store_true')
args = parser.parse_args()
args.manual_seed = 0
args.dataname = 'SceneFlow'
args.lr_schedule_type = 'cosine'
args.base_batch_size = 2
args.valid_size = None
args.opt_type = 'adam'
args.momentum = 0.9
args.no_nesterov = False
args.weight_decay = 1e-4
args.label_smoothing = 0.1
args.no_decay_keys = 'bn#bias'
args.fp16_allreduce = False
args.model_init = 'he_fout'
args.validation_frequency = 1
args.print_frequency = 10
args.n_worker = 8
args.resize_scale = 0.08
args.distort_color = 'tf'
args.image_size = '128,160,192,224'
args.continuous_size = True
args.not_sync_distributed_image_size = False
args.bn_momentum = 0.1
args.bn_eps = 1e-5
args.dropout = 0.1
args.base_stage_width = 'proxyless'
args.width_mult_list = '1.0'
args.dy_conv_scaling_mode = 1
args.independent_distributed_sampling = False
args.kd_ratio = 1.0
args.kd_type = 'ce'
if args.task == 'large':
args.path = 'exp/normal'
args.dynamic_batch_size = 1
args.n_epochs = 64
args.base_lr = 1e-3
args.warmup_epochs = 0
args.warmup_lr = -1
args.ks_list = '7'
args.expand_list = '8'
args.depth_list = '4'
args.scale_list = '4'
elif args.task == 'kernel':
args.path = 'exp/normal2kernel'
args.dynamic_batch_size = 1
args.n_epochs = 25
args.base_lr = 5e-4
args.warmup_epochs = 2
args.warmup_lr = -1
args.ks_list = '3,5,7'
args.expand_list = '8'
args.depth_list = '4'
args.scale_list = '4'
elif args.task == 'depth':
args.path = 'exp/kernel2kernel_depth'
args.dynamic_batch_size = 2
args.n_epochs = 25
args.base_lr = 5e-4
args.warmup_epochs = 2
args.warmup_lr = -1
args.ks_list = '3,5,7'
args.expand_list = '8'
args.depth_list = '2,3,4'
args.scale_list = '4'
elif args.task == 'expand':
args.path = 'exp/kernel_depth2kernel_depth_width'
args.dynamic_batch_size = 4
args.n_epochs = 25
args.base_lr = 5e-4
args.warmup_epochs = 2
args.warmup_lr = -1
args.ks_list = '3,5,7'
args.expand_list = '2,4,6,8'
args.depth_list = '2,3,4'
args.scale_list = '4'
elif args.task == 'scale':
args.path = 'exp/kernel_depth_width2kernel_depth_width_scale'
args.dynamic_batch_size = 6
args.n_epochs = 25
args.base_lr = 5e-4
args.warmup_epochs = 2
args.warmup_lr = -1
args.ks_list = '3,5,7'
args.expand_list = '2,4,6,8'
args.depth_list = '2,3,4'
args.scale_list = '2,3,4'
elif args.task in ['kitti2012', 'kitti2015', 'kitti_mix', 'kitti_sf_mix']: # finetune on kitti
args.path = 'exp/%s' % args.task
args.dynamic_batch_size = 6
args.n_epochs = 400 # since the dynamic batch size leads to multiple step.
args.warmup_epochs = 0
args.warmup_lr = -1
args.ks_list = '3,5,7'
args.expand_list = '2,4,6,8'
args.depth_list = '2,3,4'
args.scale_list = '2,3,4'
args.datapath = '/datasets/'
if args.task == 'kitti2012':
args.datapath = '/datasets/kitti2012'
if args.task == 'kitti2015':
args.datapath = '/datasets/kitti2015'
args.dataname = args.task.upper()
from ofa.stereo_matching.data_providers.stereo import StereoDataProvider
StereoDataProvider.DEFAULT_PATH = args.datapath
args.lr_schedule_type = 'multistep-100-0.5'
args.base_lr = 1e-4
else:
raise NotImplementedError
if __name__ == '__main__':
torch.multiprocessing.set_start_method('spawn')
os.makedirs(args.path, exist_ok=True)
# Initialize Horovod
hvd.init()
# Pin GPU to be used to process local rank (one GPU per process)
torch.cuda.set_device(hvd.local_rank())
num_gpus = hvd.size()
torch.manual_seed(args.manual_seed)
torch.cuda.manual_seed_all(args.manual_seed)
np.random.seed(args.manual_seed)
random.seed(args.manual_seed)
# image size
args.image_size = [int(img_size) for img_size in args.image_size.split(',')]
if len(args.image_size) == 1:
args.image_size = args.image_size[0]
MyRandomResizedCrop.CONTINUOUS = args.continuous_size
MyRandomResizedCrop.SYNC_DISTRIBUTED = not args.not_sync_distributed_image_size
# build run config from args
args.lr_schedule_param = None
args.opt_param = {
'momentum': args.momentum,
'nesterov': not args.no_nesterov,
}
#args.init_lr = args.base_lr * num_gpus # linearly rescale the learning rate
args.init_lr = args.base_lr
if args.warmup_lr < 0:
args.warmup_lr = args.base_lr
args.train_batch_size = args.base_batch_size
args.test_batch_size = args.base_batch_size * 4
run_config = DistributedStereoRunConfig(**args.__dict__, num_replicas=num_gpus, rank=hvd.rank())
# print run config information
if hvd.rank() == 0:
print('Run config:')
for k, v in run_config.config.items():
print('\t%s: %s' % (k, v))
if args.dy_conv_scaling_mode == -1:
args.dy_conv_scaling_mode = None
DynamicSeparableConv2d.KERNEL_TRANSFORM_MODE = args.dy_conv_scaling_mode
# build net from args
args.width_mult_list = [float(width_mult) for width_mult in args.width_mult_list.split(',')]
args.ks_list = [int(ks) for ks in args.ks_list.split(',')]
args.expand_list = [int(e) for e in args.expand_list.split(',')]
args.depth_list = [int(d) for d in args.depth_list.split(',')]
args.scale_list = [int(d) for d in args.scale_list.split(',')]
args.width_mult_list = args.width_mult_list[0] if len(args.width_mult_list) == 1 else args.width_mult_list
net = OFAAANet(
bn_param=(args.bn_momentum, args.bn_eps),
base_stage_width=args.base_stage_width, width_mult=args.width_mult_list,
ks_list=args.ks_list, expand_ratio_list=args.expand_list, depth_list=args.depth_list, scale_list=args.scale_list
)
""" Distributed RunManager """
# Horovod: (optional) compression algorithm.
compression = hvd.Compression.fp16 if args.fp16_allreduce else hvd.Compression.none
distributed_run_manager = DistributedRunManager(
args.path, net, run_config, compression, backward_steps=args.dynamic_batch_size, is_root=(hvd.rank() == 0)
)
distributed_run_manager.save_config()
# hvd broadcast
distributed_run_manager.broadcast()
#print('Finish broadcasting.')
# training
from ofa.stereo_matching.elastic_nn.training.progressive_shrinking import validate, train
validate_func_dict = {'image_size_list': {224},
'ks_list': sorted({min(args.ks_list), max(args.ks_list)}),
'expand_ratio_list': sorted({min(args.expand_list), max(args.expand_list)}),
'depth_list': sorted({min(net.depth_list), max(net.depth_list)}),
'scale_list': sorted({min(net.scale_list), max(net.scale_list)})}
if args.task == 'large':
train(distributed_run_manager, args,
lambda _run_manager, epoch, is_test: validate(_run_manager, epoch, is_test, **validate_func_dict))
elif args.task == 'kernel':
validate_func_dict['ks_list'] = sorted(args.ks_list)
if distributed_run_manager.start_epoch == 0:
args.ofa_checkpoint_path = 'ofa_stereo_checkpoints/ofa_stereo_D4_E8_K7_S4'
load_models(distributed_run_manager, distributed_run_manager.net, args.ofa_checkpoint_path)
else:
assert args.resume
train(distributed_run_manager, args,
lambda _run_manager, epoch, is_test: validate(_run_manager, epoch, is_test, **validate_func_dict))
elif args.task == 'depth':
from ofa.stereo_matching.elastic_nn.training.progressive_shrinking import train_elastic_depth
args.ofa_checkpoint_path = 'ofa_stereo_checkpoints/ofa_stereo_D4_E8_K357_S4'
train_elastic_depth(train, distributed_run_manager, args, validate_func_dict)
elif args.task == 'expand':
from ofa.stereo_matching.elastic_nn.training.progressive_shrinking import train_elastic_expand
args.ofa_checkpoint_path = 'ofa_stereo_checkpoints/ofa_stereo_D234_E8_K357_S4'
train_elastic_expand(train, distributed_run_manager, args, validate_func_dict)
elif args.task == 'scale':
from ofa.stereo_matching.elastic_nn.training.progressive_shrinking import train_elastic_scale
args.ofa_checkpoint_path = 'ofa_stereo_checkpoints/ofa_stereo_D234_E2468_K357_S4'
train_elastic_scale(train, distributed_run_manager, args, validate_func_dict)
elif (args.task in ['kitti2012', 'kitti2015', 'kitti_mix', 'kitti_sf_mix']):
from ofa.stereo_matching.elastic_nn.training.progressive_shrinking import train_elastic_scale
if args.task in ['kitti2012', 'kitti2015']:
args.ofa_checkpoint_path = 'ofa_stereo_checkpoints/kitti_sf_mix'
else:
args.ofa_checkpoint_path = 'ofa_stereo_checkpoints/ofa_stereo_D234_E2468_K357_S234'
train_elastic_scale(train, distributed_run_manager, args, validate_func_dict)
else:
raise NotImplementedError