chore(deps): bump pillow from 10.2.0 to 10.3.0

dataquality/utils/cv.py

@@ -3,7 +3,8 @@
 from io import BytesIO
 from typing import Any, Optional
 
-from PIL import Image
+from PIL.Image import Image
+from PIL.Image import open as Image_open
 from pydantic import UUID4
 
 from dataquality import config
@@ -18,7 +19,7 @@
 
 
 def _bytes_to_img(b: bytes) -> Image:
-    return Image.open(BytesIO(b))
+    return Image_open(BytesIO(b))
 
 
 def _write_img_bytes_to_file(

dataquality/utils/cv_smart_features.py

@@ -5,7 +5,9 @@
 import vaex
 from imagededup.methods import PHash
 from multiprocess import Pool, cpu_count
-from PIL import Image, ImageFilter, ImageStat
+from PIL import ImageFilter, ImageStat
+from PIL.Image import Image
+from PIL.Image import open as Image_open
 from vaex.dataframe import DataFrame
 
 from dataquality.exceptions import GalileoWarning
@@ -19,7 +21,7 @@
 METHODS: the methods for detecting smart feature all follow the same architecture with
 a first method to quantify the anomaly (as a real number) followed by a method to
 threshold it and return a boolean (qualitive).
-For example to detect if a method is blurry we call 
+For example to detect if a method is blurry we call
     blurriness = _blurry_laplace(image_gray)
     is_blurry = _is_blurry_laplace(blurriness)
 These methods are kept separate to allow easy generalization to the use-case where we
@@ -256,7 +258,7 @@ def _is_under_exposed(
     return q_max_over <= under_exposed_max_thresh
 
 
-def _blurry_laplace(image_gray: Image.Image) -> float:
+def _blurry_laplace(image_gray: Image) -> float:
     """
     Bluriness detector method where we compute the Variance of the Laplacian.
     We use PIL to estimate the Laplacian via the 3x3 convolution filter
@@ -284,7 +286,7 @@ def _is_blurry_laplace(
     return blurriness < blurry_thresh
 
 
-def _image_content_entropy(image: Image.Image) -> float:
+def _image_content_entropy(image: Image) -> float:
     """
     Returns the entropy of the pixels on the image. A high entropy means a more complex
     image with lots of variation in the pixel values (histogram closer to being uniform)
@@ -396,7 +398,7 @@ def _is_near_duplicate(in_frame: DataFrame) -> np.ndarray:
     return np_is_near_duplicate
 
 
-def _open_and_resize(image_path: str) -> Tuple[Image.Image, Image.Image, np.ndarray]:
+def _open_and_resize(image_path: str) -> Tuple[Image, Image, np.ndarray]:
     """
     Open the image at the given path and return a triple with
     - the original image opened with PIL
@@ -406,7 +408,7 @@ def _open_and_resize(image_path: str) -> Tuple[Image.Image, Image.Image, np.ndar
     If any of the sides of the image is larger than 2**11, resize the image so that the
     largest side is now 2**11.
     """
-    image = Image.open(image_path)
+    image = Image_open(image_path)
     image_gray = image.convert(
         "L"
     )  # TODO: check if image already grey, faster to skip that ?

dataquality/utils/semantic_segmentation/metrics.py

@@ -4,7 +4,8 @@
 
 import numpy as np
 import torch
-from PIL import Image, ImageColor
+from PIL import ImageColor
+from PIL.Image import Image, fromarray
 
 from dataquality import config
 from dataquality.clients.objectstore import ObjectStore
@@ -36,7 +37,7 @@ def calculate_and_upload_dep(
     return dep_heatmaps
 
 
-def colorize_dep_heatmap(image: Image.Image, dep_mean: int) -> Image.Image:
+def colorize_dep_heatmap(image: Image, dep_mean: int) -> Image:
     """Recolors a grayscale image to a color image based on our dep mapping"""
     color_1 = ImageColor.getrgb("#9bc33f")  # Red
     color_2 = ImageColor.getrgb("#ece113")  # Yellow
@@ -62,7 +63,7 @@ def colorize_dep_heatmap(image: Image.Image, dep_mean: int) -> Image.Image:
     colorized_image[~threshold_mask, 1] = (1 - ratio) * color_2[1] + ratio * color_3[1]
     colorized_image[~threshold_mask, 2] = (1 - ratio) * color_2[2] + ratio * color_3[2]
 
-    return Image.fromarray(colorized_image.astype(np.uint8))
+    return fromarray(colorized_image.astype(np.uint8))
 
 
 def calculate_dep_heatmaps(
@@ -166,7 +167,7 @@ def dep_heatmap_to_img(dep_heatmap: np.ndarray) -> Image:
     # Scale the array values to the range [0, 255]
     dep_heatmap = (dep_heatmap * 255).astype(np.uint8)
     # Create a PIL Image object from the numpy array as grey-scale
-    img = Image.fromarray(dep_heatmap, mode="L")
+    img = fromarray(dep_heatmap, mode="L")
     if img.size[0] > MAX_DEP_HEATMAP_SIZE or img.size[1] > MAX_DEP_HEATMAP_SIZE:
         img = img.resize((MAX_DEP_HEATMAP_SIZE, MAX_DEP_HEATMAP_SIZE))
     return img

docs/cv/coco_hf_datasets.py

@@ -4,7 +4,7 @@
 import numpy as np
 import torch
 from google.cloud import storage
-from PIL import Image
+from PIL.Image import Image, Resampling
 from torchvision import transforms
 from tqdm import tqdm
 
@@ -57,7 +57,7 @@ def __init__(
         if mask_transform is None:
             mask_transform = transforms.Compose(
                 [
-                    transforms.Resize((size, size), resample=Image.NEAREST),
+                    transforms.Resize((size, size), resample=Resampling.NEAREST),
                     transforms.ToTensor(),
                 ]
             )
@@ -117,7 +117,7 @@ def __getitem__(self, idx: int) -> Dict[str, Union[torch.Tensor, int, np.ndarray
 
         # resize image and mask to given size
         unnormalized_image = image.copy().resize(
-            (self.size, self.size), resample=Image.NEAREST
+            (self.size, self.size), resample=Resampling.NEAREST
         )
         unnormalized_image = transforms.ToTensor()(unnormalized_image)
         unnormalized_image = expand_gray_channel()(unnormalized_image)

docs/cv/cv-demo-hf.py

@@ -2,7 +2,7 @@
 from io import BytesIO
 
 from datasets import load_dataset
-from PIL import Image
+from PIL.Image import Image
 from torchvision.transforms import Compose, Normalize, RandomResizedCrop, ToTensor
 from transformers import (
     AutoFeatureExtractor,

docs/cv/cv-testing-benchmark.py

@@ -2,7 +2,7 @@
 from io import BytesIO
 
 from datasets import load_dataset
-from PIL import Image
+from PIL.Image import Image
 
 food = load_dataset("sasha/dog-food")