forked from Zhengxinyang/LAS-Diffusion
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
168 lines (153 loc) · 5.27 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
import fire
import os
from network.model_trainer import DiffusionModel
from pytorch_lightning import Trainer
from pytorch_lightning.callbacks import ModelCheckpoint
from pytorch_lightning import seed_everything
from pytorch_lightning.plugins import DDPPlugin
from utils.utils import exists
from pytorch_lightning import loggers as pl_loggers
from utils.utils import ensure_directory, run, get_tensorboard_dir, find_best_epoch
from utils.shapenet_utils import snc_category_to_synth_id_all
def train_from_folder(
sdf_folder: str = "/home/D/dataset/shapenet_sdf",
sketch_folder: str = "/home/D/dataset/shapenet_edge_our_new",
data_class: str = "chair",
results_folder: str = './results',
name: str = "debug",
image_size: int = 64,
base_channels: int = 32,
optimizier: str = "adam",
attention_resolutions: str = "4, 8",
lr: float = 2e-4,
batch_size: int = 4,
with_attention: bool = True,
num_heads: int = 4,
dropout: float = 0.1,
noise_schedule: str = "linear",
kernel_size: float = 2.0,
ema_rate: float = 0.999,
save_last: bool = True,
verbose: bool = False,
training_epoch: int = 200,
in_azure: bool = False,
new: bool = True,
continue_training: bool = False,
debug: bool = False,
use_sketch_condition: bool = True,
use_text_condition: bool = False,
seed: int = 777,
save_every_epoch: int = 20,
gradient_clip_val: float = 1.,
feature_drop_out: float = 0.1,
data_augmentation: bool = False,
view_information_ratio: float = 2.0,
vit_global: bool = False,
vit_local: bool = True,
split_dataset: bool = False,
elevation_zero: bool = False,
detail_view: bool = False,
):
if not in_azure:
debug = True
else:
debug = False
data_classes = list(snc_category_to_synth_id_all.keys())
data_classes.extend(["debug", "class_5", "class_13", "all"])
assert data_class in data_classes
results_folder = results_folder + "/" + name
ensure_directory(results_folder)
if continue_training:
new = False
if new:
run(f"rm -rf {results_folder}/*")
model_args = dict(
results_folder=results_folder,
sdf_folder=sdf_folder,
sketch_folder=sketch_folder,
data_class=data_class,
batch_size=batch_size,
lr=lr,
image_size=image_size,
noise_schedule=noise_schedule,
use_sketch_condition=use_sketch_condition,
use_text_condition=use_text_condition,
base_channels=base_channels,
optimizier=optimizier,
attention_resolutions=attention_resolutions,
with_attention=with_attention,
num_heads=num_heads,
dropout=dropout,
ema_rate=ema_rate,
verbose=verbose,
save_every_epoch=save_every_epoch,
kernel_size=kernel_size,
training_epoch=training_epoch,
gradient_clip_val=gradient_clip_val,
debug=debug,
image_feature_drop_out=feature_drop_out,
view_information_ratio=view_information_ratio,
data_augmentation=data_augmentation,
vit_global=vit_global,
vit_local=vit_local,
split_dataset=split_dataset,
elevation_zero=elevation_zero,
detail_view=detail_view
)
seed_everything(seed)
model = DiffusionModel(**model_args)
if in_azure:
try:
log_dir = get_tensorboard_dir()
except Exception as e:
log_dir = results_folder
else:
log_dir = results_folder
tb_logger = pl_loggers.TensorBoardLogger(
save_dir=log_dir,
version=None,
name='logs',
default_hp_metric=False
)
checkpoint_callback = ModelCheckpoint(
monitor="current_epoch",
dirpath=results_folder,
filename="{epoch:02d}",
save_top_k=10,
save_last=save_last,
every_n_epochs=save_every_epoch,
mode="max",
)
last_epoch = find_best_epoch(results_folder)
if os.path.exists(os.path.join(results_folder, "last.ckpt")):
last_ckpt = "last.ckpt"
else:
if exists(last_epoch):
last_ckpt = f"epoch={last_epoch:02d}.ckpt"
else:
last_ckpt = "last.ckpt"
find_unused_parameters = False
if in_azure:
trainer = Trainer(devices=-1,
accelerator="gpu",
strategy=DDPPlugin(
find_unused_parameters=find_unused_parameters),
logger=tb_logger,
max_epochs=training_epoch,
log_every_n_steps=10,
callbacks=[checkpoint_callback])
else:
trainer = Trainer(devices=-1,
accelerator="gpu",
strategy=DDPPlugin(
find_unused_parameters=find_unused_parameters),
logger=tb_logger,
max_epochs=training_epoch,
log_every_n_steps=1,
callbacks=[checkpoint_callback])
if continue_training and os.path.exists(os.path.join(results_folder, last_ckpt)):
trainer.fit(model, ckpt_path=os.path.join(results_folder, last_ckpt))
else:
trainer.fit(model)
if __name__ == '__main__':
fire.Fire(train_from_folder)