ToTransform before Normalize causes Tensor no attribute astype Error

[sarmientoj24]

This is my albumentations transform. Before, this was Normalize --> ToTensor.
Changing the order (which I think is the right order) produces an error.

def get_transforms(phase, mean, std):
    list_transforms = []
    if phase == "train":
        list_transforms.extend(
            [
                HorizontalFlip(p=0.2),
                ShiftScaleRotate(
                    shift_limit=0,  # no resizing
                    scale_limit=0.1,
                    rotate_limit=10, # rotate
                    p=0.5,
                    border_mode=cv2.BORDER_CONSTANT
                )
                # albu.RandomRotate90(),
                # albu.Cutout(),
                # albu.RandomBrightnessContrast(
                #     brightness_limit=0.2, contrast_limit=0.2, p=0.3
                # ),
                # # albu.GridDistortion(p=0.3),
                # albu.HueSaturationValue(p=0.2),
                # albu.RandomContrast(p=0.2),
                # albu.MedianBlur(p=0.2)
                # Resize(320, 480),
            ]
        )
    list_transforms.extend(
        [
            ToTensor(),
            Normalize(mean=mean, std=std, p=1),
            
        ]
    )

    list_trfms = Compose(list_transforms)
    return list_trfms

When loading using DataLoader, it generates an error

     90     denominator = np.reciprocal(std, dtype=np.float32)
     91 
---> 92     img = img.astype(np.float32)
     93     img -= mean
     94     img *= denominator

AttributeError: 'Tensor' object has no attribute 'astype'

[ternaus]

This is the expected behavior.

Normalize expects numpy arrays, and ToTensor returns torch.Tensor

[sarmientoj24]

what should be done in order to do it?

[ternaus]

Can you apply ToTensor after Normalize ?

[sarmientoj24]

Yes. But is that fine?

[sarmientoj24]

Now I am having this error when I use ShiftScale Rotate

Error

error                                     Traceback (most recent call last)
<ipython-input-164-6ec0d9d0ef8e> in <module>()
      1 fake_loader = DataLoader(train_dataset, batch_size=1,
      2                           pin_memory=True, shuffle=True)
----> 3 batch = next(iter(fake_loader)) # get a batch from the dataloader
      4 images, masks = batch

10 frames
/usr/local/lib/python3.6/dist-packages/albumentations/augmentations/functional.py in shift_scale_rotate(img, angle, scale, dx, dy, interpolation, border_mode)
    143     matrix[0, 2] += dx * width
    144     matrix[1, 2] += dy * height
--> 145     img = cv2.warpAffine(img, matrix, (width, height), flags=interpolation, borderMode=border_mode)
    146     return img
    147 

error: OpenCV(4.1.2) /io/opencv/modules/imgproc/src/imgwarp.cpp:2594: error: (-215:Assertion failed) src.cols > 0 && src.rows > 0 in function 'warpAffine'

My DataLoader/Dataset

class DPWHDataset(Dataset):
  def __init__(self, mean=None, std=None, phase=None, dataset=None):
    self.data = dataset
    self.mean = mean
    self.std = std
    self.phase = phase
    self.transforms = get_transforms(phase, mean, std)
  
  def __len__(self):
    return len(self.data)

  def __getitem__(self, idx):
    image_name = self.data[idx]

    image_path = image_prefix + image_name + ".jpg"
    mask_path = binary_mask_prefix + image_name + "_mask.png"
    print(mask_path)

    mask = cv2.imread(mask_path, 0)

    # image = np.array(Image.open(image_path))
    # mask = np.array(Image.open(mask_path))
    image = cv2.imread(image_path)
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    
    mask = create_channel_mask(mask)
    augmented = self.transforms(image=image, mask=mask)
    # image = augmented['image']
    # mask = augmented['mask']

    image = torchvision.transforms.ToTensor()(image)
    image = torchvision.transforms.Normalize(mean=self.mean, std=self.std)(image)
    mask = torchvision.transforms.ToTensor()(mask).permute(1, 2, 0)
    return image, mask

[Dipet]

Can you give reproducible example? Looks like mask or image is empty image.

[sarmientoj24]

not really an empty image. how do i give reproducible example?

[Dipet]

This error indicates that your image has 0 height or width.

error: OpenCV(4.1.2) /io/opencv/modules/imgproc/src/imgwarp.cpp:2594: error: (-215:Assertion failed) src.cols > 0 && src.rows > 0 in function 'warpAffine'

We need sample code with image and mask on which you receive these errors.

[sarmientoj24]

DataLoader

from PIL import Image
import torchvision

image_prefix = "/content/segmentation_dataset/new_img/"
color_mask_prefix = "/content/segmentation_dataset/new_color_mask/"
binary_mask_prefix = "/content/segmentation_dataset/mask/"


class DPWHDataset(Dataset):
  def __init__(self, mean=None, std=None, phase=None, dataset=None):
    self.data = dataset
    self.mean = mean
    self.std = std
    self.phase = phase
    self.transforms = get_transforms(phase, mean, std)
  
  def __len__(self):
    return len(self.data)

  def __getitem__(self, idx):
    image_name = self.data[idx]

    image_path = image_prefix + image_name + ".jpg"
    mask_path = binary_mask_prefix + image_name + "_mask.png"
    # print(mask_path)

    mask = cv2.imread(mask_path, 0)

    # image = np.array(Image.open(image_path))
    # mask = np.array(Image.open(mask_path))
    image = cv2.imread(image_path)
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    
    mask = create_channel_mask(mask)

    augmented = self.transforms(image=image, mask=mask)
    image = augmented['image']
    mask = augmented['mask']

    image = torchvision.transforms.ToTensor()(image)
    image = torchvision.transforms.Normalize(mean=self.mean, std=self.std)(image)
    mask = torchvision.transforms.ToTensor()(mask).permute(1, 2, 0)
    return image, mask

Creator of n_channel_binary_mask

def create_channel_mask(mask):
  mask_palette = [1, 2, 3, 4, 5]
  final_mask = []
  for palette in mask_palette:
    new_mask = np.zeros((mask.shape[0], mask.shape[1]), dtype=np.float32)
    new_mask[mask == palette] = 1
    final_mask.append(new_mask)

  final_mask = np.asarray(final_mask)
  return final_mask

Albumentations

import albumentations as albu

def get_transforms(phase, mean, std):
    list_transforms = []
    if phase == "train":
        list_transforms.extend(
            [
                # HorizontalFlip(p=0.5),
                # ShiftScaleRotate(
                #     shift_limit=0,  # no resizing
                #     scale_limit=0.1,
                #     rotate_limit=10, # rotate
                #     p=0.8,
                #     border_mode=cv2.BORDER_CONSTANT
                # ),
                # # albu.RandomRotate90(),
                # albu.Cutout(),
                albu.RandomBrightnessContrast(
                    brightness_limit=0.2, contrast_limit=0.2, p=0.2
                ),
                # # albu.GridDistortion(p=0.3),
                albu.HueSaturationValue(p=0.2),
                albu.RandomContrast(p=0.2),
                albu.MedianBlur(p=0.2)
            ]
        )
    # list_transforms.extend(
    #     [
    #         Resize(600, 900, interpolation=cv2.INTER_NEAREST)
    #         # Normalize(mean=mean, std=std, p=1),
    #         # ToTensor(),
    #     ]
    # )

    list_trfms = Compose(list_transforms)
    return list_trfms

Sample Mask image

[sarmientoj24]

Also, it seems like using flipping doesnt work on my mask properly


It looks wrongly flipped

[Dipet]

Looks like all works fine

import requests
import matplotlib.pyplot as plt
import albumentations as A
import cv2
import numpy as np

response = requests.get("https://raw.githubusercontent.com/albumentations-team/albumentations_examples/master/notebooks/images/parrot.jpg")
img = cv2.imdecode(np.frombuffer(response.content, dtype=np.uint8), cv2.IMREAD_COLOR)
mask = np.zeros(img.shape[:2], dtype=np.float32)
mask[200:1000, 200:700] = 1

transforms = A.Compose([
    A.HorizontalFlip(p=1),
    A.ShiftScaleRotate(
        shift_limit=0,  # no resizing
        scale_limit=0.1,
        rotate_limit=10, # rotate
        p=1,
        border_mode=cv2.BORDER_CONSTANT
    ),
    A.RandomRotate90(p=1),
    A.Cutout(p=1),
    A.RandomBrightnessContrast(
        brightness_limit=0.2, contrast_limit=0.2, p=1
    ),
    A.GridDistortion(p=1),
    A.HueSaturationValue(p=1),
    A.RandomContrast(p=1),
    A.MedianBlur(p=1),
    A.Normalize(),
])

res = transforms(image=img, mask=mask)


plt.subplot(221)
plt.imshow(img)
plt.subplot(222)
plt.imshow(mask, cmap="gray")

plt.subplot(223)
plt.imshow(res["image"])
plt.subplot(224)
plt.imshow(res["mask"], cmap="gray")


plt.show()