-
Notifications
You must be signed in to change notification settings - Fork 6
/
dataloader.py
229 lines (179 loc) · 6.23 KB
/
dataloader.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
"""
Dataloaders
"""
# Libs
from dataclasses import dataclass
import logging
from typing import TYPE_CHECKING
from sacred import Experiment
# Torch modules
from torch.utils.data import DataLoader, Dataset
import torch
import torch.distributed as dist
import cv2
from config import initialise
import random
if TYPE_CHECKING:
from utils.typing_alias import *
ex = Experiment("data")
ex = initialise(ex)
@dataclass
class Data:
train_loader: DataLoader
val_loader: DataLoader
test_loader: DataLoader
class OLEDDataset(Dataset):
"""
Assume folders have images one level deep
"""
def __init__(
self,
args,
mode: str = "train",
max_len: int = None,
is_local_rank_0: bool = True,
):
super(OLEDDataset, self).__init__()
assert mode in ["train", "val", "test"]
self.mode = mode
self.args = args
if self.mode == "train":
self.source_dir = args.train_source_dir
self.target_dir = args.train_target_dir
elif self.mode == "val":
self.source_dir = args.val_source_dir
self.target_dir = args.val_target_dir
elif self.mode == "test":
self.source_dir = args.test_source_dir
self.target_dir = None
self.max_len = max_len
self.source_paths, self.target_paths = self._load_dataset()
if is_local_rank_0:
logging.info(
f"{mode.capitalize()} Set | Source Dir: {self.source_dir} | Target Dir: {self.target_dir}"
)
self.is_local_rank_0 = is_local_rank_0
def _load_dataset(self, glob_str="*.png") -> "Union[List,List]":
source_paths = list(self.source_dir.glob(glob_str))[: self.max_len]
if self.target_dir:
target_paths = [self.target_dir / file.name for file in source_paths]
else:
target_paths = []
return source_paths, target_paths
def __len__(self):
return len(self.source_paths)
def __getitem__(self, index):
source_path = self.source_paths[index]
if self.mode == "train":
target_path = self.target_paths[index]
source = cv2.imread(str(source_path))[:, :, ::-1] / 255.0
target = cv2.imread(str(target_path))[:, :, ::-1] / 255.0
# Data augmentation
if self.args.do_augment:
# Vertical flip
if random.random() < 0.25:
source = source[::-1]
target = target[::-1]
# Horz flip
if random.random() < 0.25:
source = source[:, ::-1]
target = target[:, ::-1]
# 180 rotate
if random.random() < 0.25:
source = cv2.rotate(source, cv2.ROTATE_180)
target = cv2.rotate(target, cv2.ROTATE_180)
elif self.mode == "val":
target_path = self.target_paths[index]
source = cv2.imread(str(source_path))[:, :, ::-1] / 255.0
target = cv2.imread(str(target_path))[:, :, ::-1] / 255.0
elif self.mode == "test":
source = cv2.imread(str(source_path))[:, :, ::-1] / 255.0
source = torch.tensor(source.copy()).float().permute(2, 0, 1)
source = (source - 0.5) * 2
if self.mode in ["train", "val"]:
target = torch.tensor(target.copy()).float().permute(2, 0, 1)
target = (target - 0.5) * 2
return (source, target, source_path.name)
else:
return (source, source_path.name)
def get_dataloaders(args, is_local_rank_0: bool = True):
"""
Get dataloaders for train and val
Returns:
:data
"""
train_dataset = OLEDDataset(args, mode="train", is_local_rank_0=is_local_rank_0)
val_dataset = OLEDDataset(args, mode="val", is_local_rank_0=is_local_rank_0)
test_dataset = OLEDDataset(args, mode="test", is_local_rank_0=is_local_rank_0)
if is_local_rank_0:
logging.info(
f"Dataset Train: {len(train_dataset)} Val: {len(val_dataset)} Test: {len(test_dataset)}"
)
train_loader = None
val_loader = None
test_loader = None
if len(train_dataset):
if args.distdataparallel:
train_sampler = torch.utils.data.distributed.DistributedSampler(
train_dataset, num_replicas=dist.get_world_size(), shuffle=True
)
shuffle = False
else:
train_sampler = None
shuffle = True
train_loader = DataLoader(
train_dataset,
batch_size=args.batch_size,
shuffle=shuffle,
num_workers=args.num_threads,
pin_memory=False,
drop_last=True,
sampler=train_sampler,
)
if len(val_dataset):
if args.distdataparallel:
val_sampler = torch.utils.data.distributed.DistributedSampler(
val_dataset, num_replicas=dist.get_world_size(), shuffle=True
)
shuffle = False
else:
val_sampler = None
shuffle = True
val_loader = DataLoader(
val_dataset,
batch_size=args.batch_size,
shuffle=shuffle,
num_workers=0,
pin_memory=False,
drop_last=True,
sampler=val_sampler,
)
if len(test_dataset):
if args.distdataparallel:
test_sampler = torch.utils.data.distributed.DistributedSampler(
test_dataset, num_replicas=dist.get_world_size(), shuffle=True
)
shuffle = False
else:
test_sampler = None
shuffle = True
test_loader = DataLoader(
test_dataset,
batch_size=args.batch_size,
shuffle=shuffle,
num_workers=0,
pin_memory=False,
drop_last=True,
sampler=test_sampler,
)
return Data(
train_loader=train_loader, val_loader=val_loader, test_loader=test_loader
)
@ex.automain
def main(_run):
from tqdm import tqdm
from utils.tupperware import tupperware
args = tupperware(_run.config)
data = get_dataloaders(args)
for batch in tqdm(data.train_loader.dataset):
pass