Something weird with ImageFolder

1

I am using Imagefolder for classification task. My dataset is

animal_train
|-cat
     |-11.png
     |-12.png
|-dog
    |-21.png
    |-22.png
...
animal_val
|-cat
     |-11s.png
     |-12s.png
|-dog
    |-21s.png
    |-22s.png

I used code


transform = transforms.Compose(
    [transforms.RandomHorizontalFlip(),
    transforms.Resize(224),
    transforms.RandomCrop(112),        
    transforms.ToTensor(),
    transforms.Normalize(mean=(0.5,0.5,0.5),std=(0.5,0.5,0.5))])

train_loader = torch.utils.data.DataLoader(ImageFolder('./animal_train', transform=transform), 
                                                batch_size=16, shuffle=True, num_workers=4)
val_loader = torch.utils.data.DataLoader(ImageFolder('./animal_val', transform=transform), 
                                                batch_size=4, shuffle=False, num_workers=4)

I got very high accuracy for training but validation still zero. What is happening in my code?

2

If you are dealing with two classes, the random accuracy should be around 0.5 instead of a perfect 0 (flipping the predictions would yield a perfect score of 1.0).

Could you post the code to calculate the validation predictions and accuracy so that we could have a look?