-
Notifications
You must be signed in to change notification settings - Fork 2
/
dataset.py
26 lines (22 loc) · 888 Bytes
/
dataset.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
import pandas as pd
from PIL import Image
from torch.utils.data import Dataset
class ImageNetteDataset(Dataset):
def __init__(self, root, split='train', transform=None):
self.split = (split != 'train')
self.root = root
self.transform = transform
self.images = pd.read_csv(
root + '/noisy_imagenette.csv'
)[['path', 'noisy_labels_0', 'is_valid']]
self.images = self.images[self.images['is_valid'] == self.split]
self.images['noisy_labels_0'] = pd.Categorical(
self.images['noisy_labels_0']).codes
def __len__(self):
return len(self.images)
def __getitem__(self, idx):
item = self.images.iloc[idx]
img = Image.open(self.root + '/' + item['path']).convert('RGB')
if self.transform:
img = self.transform(img)
return img, item['noisy_labels_0']