Output and Broadcast shape mismatch in MNIST, torchvision

Question

I am getting following error when using MNIST dataset in Torchvision

RuntimeError: output with shape [1, 28, 28] doesn't match the broadcast shape [3, 28, 28]

Here is my code:

import torch
from torchvision import datasets, transforms

transform = transforms.Compose([transforms.ToTensor(),
                            transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
                          ])
trainset = datasets.MNIST('~/.pytorch/MNIST_data/', download=True, train=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=64, shuffle=True)
images, labels = next(iter(trainloader))

`MNIST` dataset has only 1 channel. You need to change `transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))` (it is for 3 channels) — Mirodil, Mar 12 '19 at 14:58

score 44 · Accepted Answer · answered Mar 12 '19 at 14:57

44

The error is due to color vs grayscale on the dataset, the dataset is grayscale.

I fixed it by changing transform to

transform = transforms.Compose([transforms.ToTensor(),
  transforms.Normalize((0.5,), (0.5,))
])

answered Mar 12 '19 at 14:57

Jibin Mathew

4,816
4
40
68

Thanks it worked. Can you just briefly explain your solution. Why did this work? – Anuvrat Tiku Apr 14 '19 at 07:16
6

transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)), 3 values means 3 channels, however for mnist there is only 1 channel, hence transforms.Normalize((0.5,), (0.5,)) – Jibin Mathew Apr 14 '19 at 07:38
Thanks Jibin, that makes sense – Anuvrat Tiku Apr 15 '19 at 16:10

Output and Broadcast shape mismatch in MNIST, torchvision

1 Answers1

Linked