Transforms for greyscale to RGB

1

hi, i have grayscale images of shape (1,48,48) i want to convert them into RGB image. i want to use pretrained model but my images are in greyscale and my dataset is csv which contains pixel values . please suggest me some ideas
@ptrblck @tom

2

img = img.expand(3, -1, -1) might be a start. But I’m quite confident that there are many other helpful people around here could have helped you with it.

Best regards

Thomas

3

my images are in numpy array loaded from csv file

train _X shape: (28709, 1, 48, 48) train _Y shape: (28709,)
val _X shape: (3589, 1, 48, 48) val _Y shape: (3589,)
test _X shape: (3589, 1, 48, 48) test _Y shape: (3589,)

i tried train_X[0].expand(3,-1,-1) and it dint worked .

i created a custom dataset and trying to apply transforms but transforms.ToPILImage is not accepting my image . so i thought to convert into RGB but dont know how to do it

class MyDataset(Dataset):
def __init__(self, data, targets, transform=None):
    self.data = data
    self.targets = torch.LongTensor(targets)
    self.transform = transform

def __getitem__(self, index):
    x = self.data[index]
    y = self.targets[index]
    if self.transform:
        x = Image.fromarray(self.data[index].astype(np.uint8).transpose(1,2,0)).convert('RGB')
        x = self.transform(x)
    return x, y

def __len__(self):
    return len(self.data)

transform = transforms.Compose([transforms.ToPILImage(),
                            transforms.Resize(256),
                            transforms.RandomHorizontalFlip(),
                            transforms.RandomVerticalFlip(),
                            transforms.RandomRotation(15),
                            transforms.CenterCrop(224),
                            transforms.ToTensor(),
                            transforms.Normalize([0.5, 0.5, 0.5],
                                                [0.5, 0.5, 0.5])])
train_data= MyDataset(train_X, train_Y, transform=transform)
valid_data = MyDataset(val_X, val_Y, transform=transform)
test_data = MyDataset(test_X, test_Y, transform=transform)

This is thowring an error TypeError: Cannot handle this data type: (1, 1, 1), |u1

4

train_X_rgb = train_X.expand(-1, 3,-1,-1)
now you have three channels (RGB) with the value of the first.

5

it didnt worked, got
error :The expanded size of the tensor (-1) isn’t allowed in a leading, non-existing dimension 0

6

If your train_X had shape (28709, 1, 48, 48) as you said, you would be able to call .expand(-1, 3, -1, -1) on it to get (28709, 3, 48, 48). But you can also grab the general idea of expanding the size-1 channel dimension to three and apply it to any other shape.

Best regards

Thomas

7

thnaks for your help
hi i solved it using repeat(3,1,1) on each image

8

murali_perumalla, how did you use repeat(3,1,1) on each image? can you please explain with example?

9

hi, sorry for late reply . i have created a dataset class and i used .repeat function in that class to convert every image into RGB.

class MyDataset(Dataset):

    def __init__(self, data, targets, transform=None):

        self.data = data

        self.targets = torch.LongTensor(targets)

        self.transform = transform

    def __getitem__(self, index):

        x = self.data[index]

        x = torch.from_numpy(x)

        y = self.targets[index]

        if self.transform:

            x=x.repeat(3,1,1)

            x = self.transform(x)

        return x, y

    def __len__(self):

        return len(self.data)