trying to import png images to torchvision

Question

I am attempting to import images for use with torch and torchvision. But I am receiving this error:

TypeError: Caught TypeError in DataLoader worker process 0.
Original Traceback (most recent call last):
  File "c:\python38\lib\site-packages\torch\utils\data\_utils\worker.py", line 178, in _worker_loop
    data = fetcher.fetch(index)
  File "c:\python38\lib\site-packages\torch\utils\data\_utils\fetch.py", line 47, in fetch
    return self.collate_fn(data)
  File "c:\python38\lib\site-packages\torch\utils\data\_utils\collate.py", line 79, in default_collate
    return [default_collate(samples) for samples in transposed]
  File "c:\python38\lib\site-packages\torch\utils\data\_utils\collate.py", line 79, in <listcomp>
    return [default_collate(samples) for samples in transposed]
  File "c:\python38\lib\site-packages\torch\utils\data\_utils\collate.py", line 81, in default_collate
    raise TypeError(default_collate_err_msg_format.format(elem_type))
TypeError: default_collate: batch must contain tensors, numpy arrays, numbers, dicts or lists; found <class 'PIL.Image.Image'>

Based on this post, I am converting them to Tensor:

https://discuss.pytorch.org/t/typeerror-default-collate-batch-must-contain-tensors-numpy-arrays-numbers-dicts-or-lists-found-class-imageio-core-util-array/62667

Here is my code:

import torch
import torchvision
import torchvision.transforms
from torchvision import datasets, transforms

transform = transforms.Compose([
    transforms.Resize(256),
    transforms.ToTensor()
])

dataset = torchvision.datasets.ImageFolder('datasets')

dataloader = torch.utils.data.DataLoader(dataset,
                                          batch_size=16,
                                          shuffle=True,
                                          num_workers=12)

tensor_dataset = []

for i, data in enumerate(dataloader, 0):
    Tensor = torch.tensor(data)
    tensor_dataset.append(Tensor.flatten)

The first last part is from https://github.com/TerragonDE/PyTorch but I have had no success. The data I am trying to load is from here:

http://www.cvlibs.net/datasets/kitti/

How can I solve this?

UPDATE:

Thanks @trialNerror, but now I am getting this error:

ValueError                                Traceback (most recent call last)
<ipython-input-6-aa72392b67e8> in <module>
      1 for i, data in enumerate(dataloader, 0):
----> 2     Tensor = torch.tensor(data)
      3     tensor_dataset.append(Tensor.flatten)

ValueError: only one element tensors can be converted to Python scalars

This is what I have found so far but am not sure how to apply it:

https://discuss.pytorch.org/t/pytorch-autograd-grad-only-one-element-tensors-can-be-converted-to-python-scalars/56681

UPDATE 2:

The reason why I didn't end up using the dataloader is because I end up getting this error:

num_epochs = 10
loss_values = list()

for epoch in range(1, num_epochs):
    for i, data in enumerate(train_array, 0):     
        outputs = model(data.unsqueeze(0))
        loss = criterion(outputs,data.unsqueeze(0))
        
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
    
        print('Epoch - %d, loss - %0.5f '%(epoch, loss.item()))
        loss_values.append(loss.item())

torch.Size([1, 16, 198, 660]) torch.Size([1, 32, 97, 328]) torch.Size([1, 1018112])

RuntimeError                              Traceback (most recent call last)
<ipython-input-106-5e6fa86df079> in <module>
      4 for epoch in range(1, num_epochs):
      5     for i, data in enumerate(train_array, 0):
----> 6         outputs = model(data.unsqueeze(0))
      7         loss = criterion(outputs,data.unsqueeze(0))
      8 

c:\python38\lib\site-packages\torch\nn\modules\module.py in __call__(self, *input, **kwargs)
    548             result = self._slow_forward(*input, **kwargs)
    549         else:
--> 550             result = self.forward(*input, **kwargs)
    551         for hook in self._forward_hooks.values():
    552             hook_result = hook(self, input, result)

<ipython-input-90-467a3f84a03f> in forward(self, x)
     29         print(out.shape)
     30 
---> 31         out = self.fc(out)
     32         print(out.shape)
     33 

c:\python38\lib\site-packages\torch\nn\modules\module.py in __call__(self, *input, **kwargs)
    548             result = self._slow_forward(*input, **kwargs)
    549         else:
--> 550             result = self.forward(*input, **kwargs)
    551         for hook in self._forward_hooks.values():
    552             hook_result = hook(self, input, result)

c:\python38\lib\site-packages\torch\nn\modules\linear.py in forward(self, input)
     85 
     86     def forward(self, input):
---> 87         return F.linear(input, self.weight, self.bias)
     88 
     89     def extra_repr(self):

c:\python38\lib\site-packages\torch\nn\functional.py in linear(input, weight, bias)
   1608     if input.dim() == 2 and bias is not None:
   1609         # fused op is marginally faster
-> 1610         ret = torch.addmm(bias, input, weight.t())
   1611     else:
   1612         output = input.matmul(weight.t())

RuntimeError: size mismatch, m1: [1 x 1018112], m2: [512 x 10] at C:\w\b\windows\pytorch\aten\src\TH/generic/THTensorMath.cpp:41

I realize that if you have m1: [a * b] and m2: [c * d] then b and c have to be the same value, but I am not sure, what is the best way to resize my images?

Answer 1

Please note that I had wanted to automatically load all PNG images in a directory as pytorch tensors. I had previously looked at posts like this (and many other web pages):

Loading a huge dataset batch-wise to train pytorch

But instead I ended up using this where I am using Image.open one by one on all the images in the directory instead of the torch DataLoader:

import numpy as np
from PIL import Image
import matplotlib.pyplot as plt

image = Image.open("datasets/image_02/data/my_image.png").convert('RGB')

import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision import transforms as tf

transforms = tf.Compose([tf.Resize(400), 
                        tf.ToTensor()])

img_tensor = transforms(image)

Answer 2

Your transform variable is unused, it should be passed to the Dataset constructor:

`dataset = torchvision.datasets.ImageFolder('datasets', transform=transform)`

Because of that, the ToTensor is never applied to your data, and thus they remain PIL images, not tensors.