torchvision.datasets.ImageFolder

ImageFolder

CLASStorchvision.datasets.ImageFolder(root, transform=None, target_transform=None, loader=<function default_loader>, is_valid_file=None) [SOURCE]

A generic data loader where the images are arranged in this way:

root/dog/xxx.png
root/dog/xxy.png
root/dog/xxz.png

root/cat/123.png
root/cat/nsdf3.png
root/cat/asd932_.png

Parameters

• root (string) – Root directory path.

• transform (callable, optional) – A function/transform that takes in an PIL image and returns a transformed version. E.g, transforms.RandomCrop

• target_transform (callable, optional) – A function/transform that takes in the target and transforms it.

• loader (callable, optional) – A function to load an image given its path.

• is_valid_file – A function that takes path of an Image file and check if the file is a valid file (used to check of corrupt files)

__getitem__(index)

Parameters

index (python:int) – Index

Returns

(sample, target) where target is class_index of the target class.

Return type

tuple

Solution to Return Image path

---

make original class of image folder to return image path

replace "torchvision.datasets.ImageFolder" to original ImageFolder to return image path.

below example

import torch
import torchvision
from torchvision import datasets, transforms

class MyImageFolder(datasets.ImageFolder):
    def __getitem__(self, index):
        return super(MyImageFolder, self).__getitem__(index), self.imgs[index]


    # transform 
transform = transforms.Compose(
    [transforms.ToTensor(),
     transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5))])
# create test loader
testset = MyImageFolder(root='/image_folder/test', 
                        transform=transform)

testloader = torch.utils.data.DataLoader(testset, batch_size=4, shuffle=False, num_workers=2)

class_names = ('REAL', 'FAKE')
device = 'cuda' if torch.cuda.is_available() else 'cpu'

Check By Code

---

test whether image path is returned now. 

# check if path can print
for i, data in enumerate(testloader):
(images,labels), (path,_) = data
images, labels = images.to(device), labels.to(device)
    print(path, "\n")


('/mtcnn_detect_resized/test/REAL/dmmvuaikkv.png', '/mtcnn_detect_resized/test/REAL/dnmowthjcj.png', '/mtcnn_detect_resized/test/REAL/doniqevxeg.png', '/mtcnn_detect_resized/test/REAL/dozjwhnedd.png') 

('/mtcnn_detect_resized/test/REAL/dpevefkefv.png', '/mtcnn_detect_resized/test/REAL/dpmgoiwhuf.png', '/mtcnn_detect_resized/test/REAL/dsnxgrfdmd.png', '/mtcnn_detect_resized/test/REAL/dtozwcapoa.png') 
...

('/mtcnn_detect_resized/test/REAL/dvkdfhrpph.png', '/mtcnn_detect_resized/test/REAL/dvtpwatuja.png', '/mtcnn_detect_resized/test/REAL/dvwpvqdflx.png', '/mtcnn_detect_resized/test/REAL/dxfdovivlw.png')

Finally

---

I recommend to check source code of ImageFolder and understand what I did now.

https://github.com/pytorch/vision/blob/master/torchvision/datasets/folder.py#L45-L74