[FIXED] pytroch data loader RuntimeError: stack expects each tensor to be equal size, but got [224, 224] at entry 0 and [224, 224, 3] at entry 1

Issue

my problem is i got two tensors in one dataset with headers image and label

when i execute simple loop all look fine unfortunetly when i make dataloader as below

training_loader = torch.utils.data.DataLoader(training_dataset, batch_size=100, shuffle=True)

and run

for i in training_loader:
print(i)

Im getting error:

RuntimeError: stack expects each tensor to be equal size, but got [224, 224] at entry 0 and [224, 224, 3] at entry 4

what can cause it and how to fix it ? thank you inadvance

Solution

It seems like one (or more) of your images is not a color image, but a gray-scale image.
Modify your loading code to force all images to be treated as color images:

img = Image.load(filename).convert('RGB')

See this answer for more details.

Answered By - Shai