When shuffle = False, the data order is not disturbed.
shuffle = True, randomized shuffle
import numpy as np
import h5py
import torch
from import DataLoader, Dataset
h5f = ('train.h5', 'w');
data1 = ([[1,2,3],
[2,5,6],
[3,5,6],
[4,5,6]])
data2 = ([[1,1,1],
[1,2,6],
[1,3,6],
[1,4,6]])
h5f.create_dataset(str('data'), data=data1)
h5f.create_dataset(str('label'), data=data2)
class Dataset(Dataset):
def __init__(self):
h5f = ('train.h5', 'r')
= h5f['data']
= h5f['label']
def __getitem__(self, index):
data = torch.from_numpy([index])
label = torch.from_numpy([index])
return data, label
def __len__(self):
assert [0] == [0], "wrong data length"
return [0]
dataset_train = Dataset()
loader_train = DataLoader(dataset=dataset_train,
batch_size=2,
shuffle = True)
for i, data in enumerate(loader_train):
train_data, label = data
print(train_data)
pytorch DataLoader usage details
Background:
I at first was on the data amplification of this piece of doubt, only saw the data transformation (), but did not see the data amplification, and then figured out, the data amplification in pytorch refers to + + multiple epochs under the joint action of the completion of the
There are a total of the following data transformations
composed = ([((448, 448)), # resize
(300), # random crop
(),
(mean=[0.5, 0.5, 0.5], # normalize
std=[0.5, 0.5, 0.5])])
Simple data-reading class that returns an image in PIL format.
class MyDataset():
def __init__(self, labels_file, root_dir, transform=None):
with open(labels_file) as csvfile:
self.labels_file = list((csvfile))
self.root_dir = root_dir
= transform
def __len__(self):
return len(self.labels_file)
def __getitem__(self, idx):
im_name = (root_dir, self.labels_file[idx][0])
im = (im_name)
if :
im = (im)
return im
Here is the main program
labels_file = "F:/test_temp/"
root_dir = "F:/test_temp"
dataset_transform = MyDataset(labels_file, root_dir, transform=composed)
dataloader = (dataset_transform, batch_size=1, shuffle=False)
"""The original dataset totaled3Pictures, in order tobatch_size=1, epochbecause of2 Show all pictures(common6sheet of paper) """
for eopch in range(2):
(figsize=(6, 6))
for ind, i in enumerate(dataloader):
a = i[0, :, :, :].numpy().transpose((1, 2, 0))
(1, 3, ind+1)
(a)
As you can see from the above images, the original image is actually re-transformed at each eopch stage, which results in an augmentation of the data.
The above is a personal experience, I hope it can give you a reference, and I hope you can support me more.