Change In dimension(CNN) using Dataloader

1

While i was creating dataloader using “CustomDatasetFromImages” class,
I got change in the dimesnion of my images from 3 to 4.

->data.shape
torch.Size([10, 4, 50, 50])

But how could this happend??
As my images r in the png format of dimension with “3”

Here is the code–
class CustomDatasetFromImages(Dataset):
def init(self,csv_path):
self.to_tensor=transform = transforms.Compose([
transforms.Resize(50),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))

])
    
    self.data_info=pd.read_csv(csv_path,header=None)
    
    self.image_arr=np.asarray(self.data_info.iloc[:,0])
    
    self.label_arr=np.asarray(self.data_info.iloc[:,1])
    
    
    self.data_len=len(self.data_info.index)
    
def __getitem__(self,index):
    single_image_name=self.image_arr[index]
    img_as_img=Image.open(single_image_name)
    
    #some_operation_arr=self.operation_arr[index]
    img_as_tensor = self.to_tensor(img_as_img)
    
    single_image_label=self.label_arr[index]
    return(img_as_tensor,single_image_label)
def __len__(self):
    return self.data_len

if name==“main”:
custom=CustomDatasetFromImages(r’/content/gdrive/My Drive/NishuOPENCV flower/flower_images/flower_labels.csv’)

data_loader = torch.utils.data.DataLoader(dataset=custom,batch_size=10,shuffle=False)

2

Usually you’ll get 4 channels, if PIL detects an alpha channel. You could slice the tensor or convert your image using img_as_img = img_as_img.convert('RGB').

3

Got it.Thanks
But what is alpha channel?

And one more thing how we can create dataloader for simpler data in stored csv file.

4

The alpha channel is usually used to create transparency as descibed here.

Usually you would implement your custom Dataset, set the path to the .csv data in its __init__, and load the samples in __getitem__. Basically what you’ve done in your first example. Do you have any trouble with your code or am I misunderstanding the question?