-
Notifications
You must be signed in to change notification settings - Fork 0
/
dataset_generator.py
48 lines (38 loc) · 1.44 KB
/
dataset_generator.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
import pandas as pd
from torch.utils.data import Dataset
from PIL import Image
import os
import numpy as np
import torch
from collections import Counter
class SignsDataset(Dataset):
def __init__(self, file, folder, signs_subset=None, transform=None):
signs = pd.read_csv(file)
if signs_subset is not None:
assert isinstance(signs_subset, (list, tuple))
mask = signs['class_number'].isin(signs_subset)
signs = signs[mask]
else:
signs_subset = np.unique(self.signs['class_number'])
self.filenames = signs['filename']
self.labels = signs['class_number']
self.transform = transform
self.folder = folder
self.mapping = {label: i for i, label in enumerate(np.unique(self.labels))}
N, K = len(self.labels), len(self.mapping)
counter = Counter()
counter.update(self.labels)
counts = [(self.mapping[label], cnt) for label, cnt in counter.items()]
self.weights = [1] * K
def __getitem__(self, i):
filename = self.filenames.iloc[i]
label = self.labels.iloc[i]
path = os.path.join(self.folder, filename)
image = Image.open(path)
if self.transform is not None:
image = self.transform(image)
return image, self.mapping[label]
def __len__(self):
return len(self.filenames)
def get_weights(self):
return torch.FloatTensor(self.weights)