-
Notifications
You must be signed in to change notification settings - Fork 1
/
main_multicore.py
100 lines (81 loc) · 4.13 KB
/
main_multicore.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
import os
import typing as t
from pathlib import Path
from easydict import EasyDict as edict
from loguru import logger
from contrastyou import CONFIG_PATH, git_hash, OPT_PATH
from contrastyou.arch import UNet
from contrastyou.configure import ConfigManager
from contrastyou.configure.yaml_parser import yaml_load
from contrastyou.losses.multicore_loss import MultiCoreKL
from contrastyou.trainer import create_save_dir
from contrastyou.utils import fix_all_seed_within_context, adding_writable_sink, extract_model_state_dict
from hook_creator import create_hook_from_config
from semi_seg.data.creator import get_data
from semi_seg.trainers.features import MulticoreTrainer
from utils import logging_configs, find_checkpoint, grouper
@logger.catch(reraise=True)
def main():
manager = ConfigManager(os.path.join(CONFIG_PATH, "base.yaml"), strict=True, verbose=False)
with manager(scope="base") as config:
# this handles input save dir with relative and absolute paths
absolute_save_dir = create_save_dir(MulticoreTrainer, config["Trainer"]["save_dir"])
if os.path.exists(absolute_save_dir):
logger.warning(f"{absolute_save_dir} exists, may overwrite the folder")
adding_writable_sink(absolute_save_dir)
logging_configs(manager, logger)
config.update({"GITHASH": git_hash})
seed = config.get("RandomSeed", 10)
logger.info(f"using seed = {seed}, saved at \"{absolute_save_dir}\"")
with fix_all_seed_within_context(seed):
worker(config, absolute_save_dir, seed)
def worker(config, absolute_save_dir, seed):
# load data setting
data_name = config.Data.name
data_opt = edict(yaml_load(Path(OPT_PATH) / (data_name + ".yaml")))
config.OPT = data_opt
model_checkpoint = config["Arch"].pop("checkpoint", None)
with fix_all_seed_within_context(seed):
config["Arch"].pop("true_num_classes", None)
true_num_classes = data_opt["num_classes"]
multiplier = config["MulticoreParameters"]["multiplier"]
model = UNet(**config["Arch"], input_dim=data_opt["input_dim"], num_classes=multiplier * true_num_classes)
config["Arch"]["true_num_classes"] = true_num_classes
criterion_name = config["MulticoreParameters"]["name"]
if criterion_name == "naive":
sup_criterion = MultiCoreKL(groups=list(grouper(range(true_num_classes * multiplier), true_num_classes)))
else:
raise RuntimeError(criterion_name)
# sup_criterion = StricterAdaptiveOverSegmentedLossWithMI(
# input_num_classes=true_num_classes * multiplier,
# output_num_classes=true_num_classes,
# device=config.Trainer.device,
# mi_weight=config["MulticoreParameters"]["mi_weight"]
# )
if model_checkpoint:
logger.info(f"loading checkpoint from {model_checkpoint}")
model.load_state_dict(extract_model_state_dict(model_checkpoint), strict=True)
total_freedom = False
labeled_loader, unlabeled_loader, val_loader, test_loader = get_data(
data_params=config["Data"], labeled_loader_params=config["LabeledLoader"],
unlabeled_loader_params=config["UnlabeledLoader"], pretrain=False, total_freedom=total_freedom)
trainer = MulticoreTrainer(
model=model, labeled_loader=labeled_loader, unlabeled_loader=unlabeled_loader,
val_loader=val_loader, test_loader=test_loader, criterion=sup_criterion, config=config,
save_dir=absolute_save_dir,
**{k: v for k, v in config["Trainer"].items() if k != "save_dir" and k != "name"}
)
# find the last.pth from the save folder.
checkpoint: t.Optional[str] = find_checkpoint(trainer.absolute_save_dir)
with fix_all_seed_within_context(seed):
hooks = create_hook_from_config(model, config, is_pretrain=False, trainer=trainer)
hook_registration = trainer.register_hook
with hook_registration(*hooks):
trainer.init()
if checkpoint:
trainer.resume_from_path(checkpoint)
return trainer.start_training()
if __name__ == '__main__':
import torch
torch.backends.cudnn.benchmark = True
main()