-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
195 lines (169 loc) · 7.99 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
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
import yaml
# 0.1 Load configfile
with open('config.yaml') as yamlfile:
config = yaml.load(yamlfile, Loader=yaml.FullLoader)
import os
import torch
import wandb
from torch.utils.data import DataLoader, DistributedSampler
import torch.multiprocessing as mp
from torchinfo import summary
from typing import Dict
from datetime import datetime
import torch.distributed as dist
from torch.nn.parallel import DistributedDataParallel as DDP
from engine import train_one_epoch, evaluate, EarlyStopping
from dataloader import StixelData
if config['mode'] == "segmentation":
import models.ConvNeXt_pretrained as model_file
from models import get_model as model_fn
from losses import StixelVoxelLoss as StixelLoss
elif config['mode'] == "classification":
import models.ConvNeXt_pretrained as model_file
from models import convnext_stixel as model_fn
from losses import StixelObjectLoss as StixelLoss
else:
raise ValueError("Invalid mode specified in config file!")
# starting time for all instances
overall_start_time = datetime.now()
def setup(rank, world_size):
os.environ['MASTER_ADDR'] = 'localhost'
os.environ['MASTER_PORT'] = '12356'
# 'gloo' for CPUs, 'nccl' for GPUs
dist.init_process_group("nccl", rank=rank, world_size=world_size)
torch.cuda.set_device(rank)
def cleanup():
dist.barrier()
dist.destroy_process_group()
def save_checkpoint(model, optimizer, epoch, loss, filename):
torch.save({
'epoch': epoch,
'model_state_dict': model.state_dict(),
'optimizer_state_dict': optimizer.state_dict(),
'loss': loss,
}, filename)
def load_checkpoint(model, optimizer, filename):
checkpoint = torch.load(filename)
model.load_state_dict(checkpoint['model_state_dict'])
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
continue_epoch = checkpoint['epoch'] + 1
last_loss = checkpoint['loss']
return continue_epoch, last_loss
def train(rank, world_size):
# torch.cuda.init()
setup(rank, world_size)
""" 1.Load data """
# Training data
tmpdir = os.getenv("TMPDIR", "")
data_dir = os.path.join(tmpdir, config['data_path'])
training_data = StixelData(data_dir=data_dir, phase='training', mode=config['mode'],
target_trans_blur=config['blur'], depth_anchors=(4, 66, config['n_cand']),
transform=True)
training_sampler = DistributedSampler(training_data, num_replicas=world_size, rank=rank)
train_dataloader = DataLoader(training_data, batch_size=config['batch_size'], pin_memory=True, drop_last=True,
sampler=training_sampler)
# Validation data
validation_data = StixelData(data_dir=data_dir, phase='validation', mode=config['mode'],
depth_anchors=(4, 66, config['n_cand']),
transform=True)
# validation_sampler = DistributedSampler(validation_data, num_replicas=world_size, rank=rank)
val_dataloader = DataLoader(validation_data, batch_size=config['batch_size'], pin_memory=True, drop_last=True)
""" 2.Define Model & Loss """
model, model_cfg = model_fn(n_candidates=config['n_cand'])
model = model.to(rank)
model = DDP(model, device_ids=[rank], find_unused_parameters=True)
# Optimizer definition
optimizer = torch.optim.AdamW(model.parameters(), lr=config['learning_rate'])
# Loss initialization
loss_weights: Dict[str, float] = {}
if config['mode'] == "segmentation":
loss_weights = config['loss_w_seg']
elif config['mode'] == "classification":
loss_weights = config['loss_w_cls']
loss_fn = StixelLoss(loss_weights)
# Load checkpoint
start_epoch = 0
if config['load_checkpoint'] is not None:
if rank == 0 and os.path.isfile(config['load_checkpoint']):
start_epoch, loss = load_checkpoint(model, optimizer, config['load_checkpoint'])
print(
f"Checkpoint {os.path.basename(config['load_checkpoint'])} loaded. Training stopped on epoch {start_epoch - 1} with loss {loss}. Training will be continued ...")
# Initialize Logger
if config['logging'] and rank == 0:
wandb_logger = wandb.init(project="StixelNExT-Pro",
config={
"learning_rate": config['learning_rate'],
"loss_name": type(loss_fn).__name__,
"loss": loss_fn.params(),
"mode": config['mode'],
"model": model_cfg,
"dataset": training_data.name,
"epochs": config['epochs'],
"rank": rank,
"batch_size": config['batch_size'],
"checkpoint": config['load_checkpoint'],
"early_stop": config['early_stop'],
"num_gpu": world_size,
"blur": config['blur']
},
job_type="training",
tags=["training"]
)
artifact = wandb.Artifact(f"{model_cfg['name']}_weights_art", type='model',
description="Automatic checkpoint pick by train/ eval loss.")
artifact.add_file(model_file.__file__)
wandb_logger.watch(model)
else:
wandb_logger = None
""" 3.Training """
# Inspect model
summary(model, (config['batch_size'], 3, 384, 1280))
# Training
early_stopping = EarlyStopping(tolerance=config['early_stop']['tol'],
min_delta=config['early_stop']['min_delta'])
best_loss = float('inf')
for epoch in range(start_epoch, config['epochs']):
print(f"\n Epoch {epoch}\n----------------------------------------------------------------")
training_sampler.set_epoch(epoch)
train_loss = train_one_epoch(train_dataloader, model, loss_fn, optimizer,
device=rank, writer=wandb_logger)
eval_loss = evaluate(val_dataloader, model, loss_fn,
device=rank, writer=wandb_logger)
# Save model
if config['logging'] and rank == 0:
saved_models_path = os.path.join('saved_models', wandb_logger.name)
os.makedirs(saved_models_path, exist_ok=True)
weights_name = f"StixelNExT-Pro_{wandb_logger.name}_{epoch}.pth"
weights_path = os.path.join(saved_models_path, weights_name)
save_checkpoint(model, optimizer, epoch, eval_loss, weights_path)
print("Saved PyTorch Model State to " + weights_path)
if eval_loss < best_loss:
best_loss = eval_loss
best_weights_path = weights_path
artifact.metadata = {
'epoch': epoch,
'train_loss': train_loss,
'eval_loss': eval_loss
}
step_time = datetime.now() - overall_start_time
print("Time elapsed: {}".format(step_time))
# early stopping
early_stopping.check_stop(eval_loss, rank)
if early_stopping.early_stop:
print("Early stopping at epoch:", epoch)
break
overall_time = datetime.now() - overall_start_time
print(f"Finished training in {str(overall_time).split('.')[0]}")
if config['logging'] and rank == 0:
artifact.metadata.update(model_cfg)
artifact.metadata.update({"mode": config['mode']})
artifact.add_file(best_weights_path)
wandb_logger.log_artifact(artifact)
wandb.finish()
cleanup()
def main():
world_size = torch.cuda.device_count()
print(f"Found {world_size} cuda devices.")
mp.spawn(train, args=(world_size,), nprocs=world_size, join=True)
if __name__ == '__main__':
main()