-
Notifications
You must be signed in to change notification settings - Fork 0
/
batch_loader.py
155 lines (128 loc) · 4.42 KB
/
batch_loader.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
import torch
import torch.nn as nn
from pathlib import Path
from PIL import Image
import numpy as np
import matplotlib.pyplot as plt
import torch
import random
from typing import List, Dict, NamedTuple
from protonet import ProtoNet
import torch.nn.functional as F
from tqdm import tqdm
def img2arr(image_path: str,
size: int = 32) -> torch.Tensor:
image_tensor = Image.open(image_path)
image_tensor = image_tensor.resize(size=(32, 32))
image_tensor = np.asarray(image_tensor)/255
image_tensor = image_tensor.transpose(-1, 0, 1)
image_tensor = (
torch.from_numpy(image_tensor)
.to(torch.float32)
)
return image_tensor
def get_images_and_labels(data_path="./BIRDS-450-FS"):
images: Dict[str, List[torch.Tensor]] = {
path.name: [img2arr(image_path)
for image_path in path.glob("*")]
for path in Path(data_path).glob("*")
}
labels: List[str] = [*images.keys()]
return images, labels
def split_labels(labels: List[str], split_ratio=0.8):
assert split_ratio <= 1 and split_ratio > 0
random.shuffle(labels)
train_labels = labels[:int(len(labels) * split_ratio)]
test_labels = labels[int(len(labels) * split_ratio):]
return (train_labels, test_labels)
class Batch(NamedTuple):
X_spt: torch.Tensor
X_qry: torch.Tensor
y_spt: torch.LongTensor
y_qry: torch.LongTensor
class SamplingResult(NamedTuple):
batch: Batch
id2cls: List[str]
class BatchLoader:
def __init__(self,
num_way: int,
num_spt: int,
device,
num_qry: int = 1,
data_path="./BIRDS-450-FS",
split_ratio=0.8) -> None:
(self.images,
self.labels) = get_images_and_labels(data_path)
(self.train_labels,
self.test_labels) = split_labels(self.labels)
self.num_way = num_way
self.num_spt = num_spt
self.device = device
self.num_qry = num_qry
self.split_ratio = split_ratio
def sample(self):
classes: List[str] = random.sample(self.labels,
self.num_way)
c, h, w = self.images[classes[0]][0].size()
X_spt: List[torch.Tensor] = []
X_qry: List[torch.Tensor] = []
y_spt: List[int] = []
y_qry: List[int] = []
for i, cls in enumerate(classes):
image_tensors = random.sample(self.images[cls],
self.num_spt + self.num_qry)
X_spt += image_tensors[:self.num_spt]
X_qry += image_tensors[self.num_spt:]
y_spt += [i] * self.num_spt
y_qry += [i] * self.num_qry
X_spt = torch.stack(X_spt).to(device=self.device)
X_qry = torch.stack(X_qry).to(device=self.device)
y_spt = torch.tensor(y_spt).to(device=self.device)
y_qry = torch.tensor(y_qry).to(device=self.device)
assert X_spt.size() == (self.num_way*self.num_spt,
c, h, w)
assert X_qry.size() == (self.num_way*self.num_qry,
c, h, w)
return SamplingResult(batch=Batch(X_spt,
X_qry,
y_spt,
y_qry),
id2cls=classes)
def visualize_batch(batch: Batch,
id2cls: List[str]):
(X_spt,
X_qry,
y_spt,
y_qry) = batch
num_way: int = torch.unique(y_spt).size(0)
num_spt: int = y_spt.size(0)//num_way
num_qry: int = y_qry.size(0)//num_way
fig, axes = plt.subplots(
ncols=num_way,
nrows=num_spt,
figsize=(num_way*3, 1 + num_spt*3,),
subplot_kw={
"xticks": [],
"yticks": []
}
)
fig.suptitle("Support Set", fontsize=20)
axes = axes.T.reshape(-1)
for i, ax in enumerate(axes):
ax.imshow(X_spt[i, :, :, :].permute(1, 2, 0))
ax.set_title(id2cls[y_spt[i]])
fig, axes = plt.subplots(
ncols=num_way,
nrows=num_qry,
figsize=(num_way*3, 1 + num_qry*3),
subplot_kw={
"xticks": [],
"yticks": []
}
)
fig.suptitle("Query Set", fontsize=20)
axes = axes.T.reshape(-1)
for i, ax in enumerate(axes):
ax.imshow(X_qry[i, :, :, :].permute(1, 2, 0))
ax.set_title(id2cls[y_qry[i]])
plt.show()