-
Notifications
You must be signed in to change notification settings - Fork 1
/
datamodule.py
201 lines (165 loc) · 6.75 KB
/
datamodule.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
import torch
import torchvision.transforms as T
from torch.utils.data import DataLoader
import kornia.augmentation as K
import lightning.pytorch as pl
from einops import rearrange
from matplotlib import colors
import matplotlib.pyplot as plt
from dfc2022 import DFC2022
from torchgeo.datasets.utils import percentile_normalization
DEFAULT_AUGS = K.AugmentationSequential(
K.RandomHorizontalFlip(p=0.5),
K.RandomVerticalFlip(p=0.5),
data_keys=["input", "mask"],
)
class DFC2022DataModule(pl.LightningDataModule):
# Stats computed in labeled train set
dem_min, dem_max = -79.18, 3020.26
dem_nodata = -99999.0
def __init__(
self,
root_dir: str,
batch_size: int = 8,
num_workers: int = 0,
train_coordinate_file_path: str = 'train_coords.txt',
training_sample_amount: int = 77133,
val_image_file_path: str = 'val_coords.txt',
patch_size: int = 256,
augmentations=DEFAULT_AUGS,
**kwargs,
):
super().__init__()
self.root_dir = root_dir
self.batch_size = batch_size
self.num_workers = num_workers
self.train_coordinate_file_path = train_coordinate_file_path
self.training_sample_amount = training_sample_amount
self.val_image_file_path = val_image_file_path
self.patch_size = patch_size
self.augmentations = augmentations
# self.random_crop = T.RandomCrop((self.patch_size, self.patch_size))
self.random_crop = K.AugmentationSequential(
K.RandomCrop((self.patch_size, self.patch_size), p=1.0, keepdim=False),
data_keys=["input", "mask"],
)
self.train_dataset = None
self.val_dataset = None
self.test_dataset = None
self.predict_dataset = None
def preprocess(self, sample):
# RGB is uint8 so divide by 255
sample["image"][:3] /= 255.0
sample["image"][-1] = (sample["image"][-1] - self.dem_min) / (
self.dem_max - self.dem_min
)
sample["image"][-1] = torch.clip(sample["image"][-1], min=0.0, max=1.0)
if "mask" in sample:
# ignore the clouds and shadows class (not used in scoring)
sample["mask"][sample["mask"] == 15] = 0
sample["mask"] = rearrange(sample["mask"], "h w -> () h w")
return sample
# this is only used to simplify the validation process
# the validation process is performed using crops of the validation images
def crop(self, sample):
sample["image"] = rearrange(sample["image"], "c h w -> () c h w")
sample["mask"] = rearrange(sample["mask"], "c h w -> () c h w")
sample["mask"] = sample["mask"].to(torch.float)
sample["image"], sample["mask"] = self.random_crop(
sample["image"], sample["mask"]
)
sample["mask"] = sample["mask"].to(torch.long)
sample["image"] = rearrange(sample["image"], "() c h w -> c h w")
sample["mask"] = rearrange(sample["mask"], "() c h w -> c h w")
return sample
def setup(self, stage=None):
train_transforms = T.Compose([self.preprocess])
val_transforms = T.Compose([self.preprocess, self.crop])
test_transforms = T.Compose([self.preprocess])
self.train_dataset = DFC2022(self.root_dir, self.train_coordinate_file_path, "train",
self.patch_size, training_sample_amount=self.training_sample_amount,
transforms=train_transforms)
# for validation, there is no patch size for the dataloader since patches are generated using the transforms
self.val_dataset = DFC2022(self.root_dir, self.val_image_file_path, "train_val",
patch_size=-1, transforms=val_transforms)
# for test, there is no patch size since images are processed entirely
self.test_dataset = DFC2022(self.root_dir, None, "test",
patch_size=-1, transforms=test_transforms)
def train_dataloader(self):
return DataLoader(
self.train_dataset,
batch_size=self.batch_size,
num_workers=self.num_workers,
shuffle=True,
)
def val_dataloader(self):
return DataLoader(
self.val_dataset,
batch_size=self.batch_size,
num_workers=self.num_workers,
shuffle=False
)
def test_dataloader(self):
return DataLoader(
self.test_dataset,
batch_size=1,
num_workers=self.num_workers,
shuffle=False,
)
def on_after_batch_transfer(self, batch, dataloader_idx):
if self.trainer.training:
if self.augmentations is not None:
batch["mask"] = batch["mask"].to(torch.float)
batch["image"], batch["mask"] = self.augmentations(
batch["image"], batch["mask"]
)
batch["mask"] = batch["mask"].to(torch.long)
batch["mask"] = rearrange(batch["mask"], "b () h w -> b h w")
return batch
def plot(
self,
sample,
show_titles=True,
suptitle=None,
):
ncols = 2
image = sample["image"][:3]
image = image.to(torch.uint8)
image = image.permute(1, 2, 0).numpy()
dem = sample["image"][-1].numpy()
dem = percentile_normalization(dem, lower=0, upper=100, axis=(0, 1))
showing_mask = "mask" in sample
showing_prediction = "prediction" in sample
cmap = colors.ListedColormap(DFC2022.colormap)
if showing_mask:
mask = sample["mask"].numpy()
ncols += 1
if showing_prediction:
pred = sample["prediction"].numpy()
ncols += 1
fig, axs = plt.subplots(nrows=1, ncols=ncols, figsize=(10, ncols * 10))
axs[0].imshow(image)
axs[0].axis("off")
axs[1].imshow(dem)
axs[1].axis("off")
if showing_mask:
axs[2].imshow(mask, cmap=cmap, interpolation="none")
axs[2].axis("off")
if showing_prediction:
axs[3].imshow(pred, cmap=cmap, interpolation="none")
axs[3].axis("off")
elif showing_prediction:
axs[2].imshow(pred, cmap=cmap, interpolation="none")
axs[2].axis("off")
if show_titles:
axs[0].set_title("Image")
axs[1].set_title("DEM")
if showing_mask:
axs[2].set_title("Ground Truth")
if showing_prediction:
axs[3].set_title("Predictions")
elif showing_prediction:
axs[2].set_title("Predictions")
if suptitle is not None:
plt.suptitle(suptitle)
return fig