Split some classes from Tiny-ImageNet dataset

1

Hello all,
I am trying to split class labels 0 to 9 of the Tiny-imagenet dataset so I tried the following code

    train_dataset = TinyImageNet('tiny-imagenet-200', 'train', transform=transform)
    train_labels_np=np.array(list(train_dataset.labels.values()))
    train_indexes=np.where((train_labels_np==0)|(train_labels_np==1)|
                            (train_labels_np==2)|(train_labels_np==3)|
                            (train_labels_np==4)|(train_labels_np==5)|
                            (train_labels_np==6)|(train_labels_np==7)|
                            (train_labels_np==8)|(train_labels_np==9))[0]
    np.random.shuffle(train_indexes)    #Shuffle   
    training_split_indexes = train_indexes[0:2048]    #select some examples
    sampler_train= torch.utils.data.SubsetRandomSampler(training_split_indexes)
 
    train_set = torch.utils.data.DataLoader(train_dataset, batch_size=args.train_batch_size, sampler=sampler_train, **kwargs)
    test_dataset = TinyImageNet('tiny-imagenet-200', 'val', transform=transform)
    test_labels_np=np.array(list(test_dataset.labels.values()))
    test_indexes=np.where((test_labels_np==0)|(test_labels_np==1)|
                            (test_labels_np==2)|(test_labels_np==3)|
                            (test_labels_np==4)|(test_labels_np==5)|
                            (test_labels_np==6)|(test_labels_np==7)|
                            (test_labels_np==8)|(test_labels_np==9))[0]
  
    sampler_test = torch.utils.data.SubsetRandomSampler(test_indexes)
    test_set = torch.utils.data.DataLoader(test_dataset, batch_size=args.test_batch_size, shuffle=False, sampler=sampler_test, **kwargs)

Error IndexError: Target 68 is out of bounds. has occurred from
loss= self.cross_entropy_loss(output, target)
in test function.

how can I handle it?
thanks.

1 Like