onnxtr 0.3.1__tar.gz → 0.3.2__tar.gz

{onnxtr-0.3.1 → onnxtr-0.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: onnxtr
-Version: 0.3.1
+Version: 0.3.2
 Summary: Onnx Text Recognition (OnnxTR): docTR Onnx-Wrapper for high-performance OCR on documents.
 Author-email: Felix Dittrich <felixdittrich92@gmail.com>
 Maintainer: Felix Dittrich

{onnxtr-0.3.1 → onnxtr-0.3.2}/onnxtr/transforms/base.py

@@ -5,8 +5,8 @@
 
 from typing import Tuple, Union
 
-import cv2
 import numpy as np
+from PIL import Image, ImageOps
 
 __all__ = ["Resize", "Normalize"]
 
@@ -17,64 +17,51 @@ class Resize:
     def __init__(
         self,
         size: Union[int, Tuple[int, int]],
-        interpolation=cv2.INTER_LINEAR,
+        interpolation=Image.Resampling.BILINEAR,
         preserve_aspect_ratio: bool = False,
         symmetric_pad: bool = False,
     ) -> None:
-        super().__init__()
-        self.size = size
+        self.size = size if isinstance(size, tuple) else (size, size)
         self.interpolation = interpolation
         self.preserve_aspect_ratio = preserve_aspect_ratio
         self.symmetric_pad = symmetric_pad
         self.output_size = size if isinstance(size, tuple) else (size, size)
 
-        if not isinstance(self.size, (int, tuple, list)):
-            raise AssertionError("size should be either a tuple, a list or an int")
+        if not isinstance(self.size, (tuple, int)):
+            raise AssertionError("size should be either a tuple or an int")
 
-    def __call__(
-        self,
-        img: np.ndarray,
-    ) -> np.ndarray:
-        if img.ndim == 3:
-            h, w = img.shape[0:2]
-        else:
-            h, w = img.shape[1:3]
-        sh, sw = self.size if isinstance(self.size, tuple) else (self.size, self.size)
+    def __call__(self, img: np.ndarray) -> np.ndarray:
+        img = (img * 255).astype(np.uint8) if img.dtype != np.uint8 else img
+        h, w = img.shape[:2] if img.ndim == 3 else img.shape[1:3]
+        sh, sw = self.size
 
-        # Calculate aspect ratio of the image
-        aspect = w / h
+        if not self.preserve_aspect_ratio:
+            return np.array(Image.fromarray(img).resize((sw, sh), resample=self.interpolation))
 
-        # Compute scaling and padding sizes
-        if self.preserve_aspect_ratio:
-            if aspect > 1:  # Horizontal image
-                new_w = sw
-                new_h = int(sw / aspect)
-            elif aspect < 1:  # Vertical image
-                new_h = sh
-                new_w = int(sh * aspect)
-            else:  # Square image
-                new_h, new_w = sh, sw
-
-            img_resized = cv2.resize(img, (new_w, new_h), interpolation=self.interpolation)
-
-            # Calculate padding
-            pad_top = max((sh - new_h) // 2, 0)
-            pad_bottom = max(sh - new_h - pad_top, 0)
-            pad_left = max((sw - new_w) // 2, 0)
-            pad_right = max(sw - new_w - pad_left, 0)
-
-            # Pad the image
-            img_resized = cv2.copyMakeBorder(  # type: ignore[call-overload]
-                img_resized, pad_top, pad_bottom, pad_left, pad_right, borderType=cv2.BORDER_CONSTANT, value=0
-            )
-
-            # Ensure the image matches the target size by resizing it again if needed
-            img_resized = cv2.resize(img_resized, (sw, sh), interpolation=self.interpolation)
+        actual_ratio = h / w
+        target_ratio = sh / sw
+
+        if target_ratio == actual_ratio:
+            return np.array(Image.fromarray(img).resize((sw, sh), resample=self.interpolation))
+
+        if actual_ratio > target_ratio:
+            tmp_size = (int(sh / actual_ratio), sh)
         else:
-            # Resize the image without preserving aspect ratio
-            img_resized = cv2.resize(img, (sw, sh), interpolation=self.interpolation)
+            tmp_size = (sw, int(sw * actual_ratio))
+
+        img_resized = Image.fromarray(img).resize(tmp_size, resample=self.interpolation)
+        pad_left = pad_top = 0
+        pad_right = sw - img_resized.width
+        pad_bottom = sh - img_resized.height
+
+        if self.symmetric_pad:
+            pad_left = pad_right // 2
+            pad_right -= pad_left
+            pad_top = pad_bottom // 2
+            pad_bottom -= pad_top
 
-        return img_resized
+        img_resized = ImageOps.expand(img_resized, (pad_left, pad_top, pad_right, pad_bottom))
+        return np.array(img_resized)
 
     def __repr__(self) -> str:
         interpolate_str = self.interpolation

{onnxtr-0.3.1 → onnxtr-0.3.2}/onnxtr/utils/fonts.py

@@ -5,14 +5,16 @@
 
 import logging
 import platform
-from typing import Optional
+from typing import Optional, Union
 
 from PIL import ImageFont
 
 __all__ = ["get_font"]
 
 
-def get_font(font_family: Optional[str] = None, font_size: int = 13) -> ImageFont.ImageFont:
+def get_font(
+    font_family: Optional[str] = None, font_size: int = 13
+) -> Union[ImageFont.FreeTypeFont, ImageFont.ImageFont]:
     """Resolves a compatible ImageFont for the system
 
     Args:
@@ -29,7 +31,7 @@ def get_font(font_family: Optional[str] = None, font_size: int = 13) -> ImageFon
         try:
             font = ImageFont.truetype("FreeMono.ttf" if platform.system() == "Linux" else "Arial.ttf", font_size)
         except OSError:  # pragma: no cover
-            font = ImageFont.load_default()
+            font = ImageFont.load_default()  # type: ignore[assignment]
             logging.warning(
                 "unable to load recommended font family. Loading default PIL font,"
                 "font size issues may be expected."

onnxtr-0.3.2/onnxtr/version.py

@@ -0,0 +1 @@
+__version__ = 'v0.3.2'

{onnxtr-0.3.1 → onnxtr-0.3.2}/onnxtr.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: onnxtr
-Version: 0.3.1
+Version: 0.3.2
 Summary: Onnx Text Recognition (OnnxTR): docTR Onnx-Wrapper for high-performance OCR on documents.
 Author-email: Felix Dittrich <felixdittrich92@gmail.com>
 Maintainer: Felix Dittrich

{onnxtr-0.3.1 → onnxtr-0.3.2}/setup.py

@@ -9,7 +9,7 @@ from pathlib import Path
 from setuptools import setup
 
 PKG_NAME = "onnxtr"
-VERSION = os.getenv("BUILD_VERSION", "0.3.1a0")
+VERSION = os.getenv("BUILD_VERSION", "0.3.2a0")
 
 
 if __name__ == "__main__":

onnxtr-0.3.1/onnxtr/version.py

@@ -1 +0,0 @@
-__version__ = 'v0.3.1'

{onnxtr-0.3.1 → onnxtr-0.3.2}/onnxtr/models/preprocessor/base.py

@@ -67,11 +67,12 @@ class PreProcessor(NestedObject):
             if x.dtype not in (np.uint8, np.float32):
                 raise TypeError("unsupported data type for numpy.ndarray")
         x = shape_translate(x, "HWC")
+
+        # Resizing
+        x = self.resize(x)
         # Data type & 255 division
         if x.dtype == np.uint8:
             x = x.astype(np.float32) / 255.0
-        # Resizing
-        x = self.resize(x)
 
         return x
 
@@ -95,13 +96,12 @@ class PreProcessor(NestedObject):
                     raise TypeError("unsupported data type for numpy.ndarray")
             x = shape_translate(x, "BHWC")
 
-            # Data type & 255 division
-            if x.dtype == np.uint8:
-                x = x.astype(np.float32) / 255.0
             # Resizing
             if (x.shape[1], x.shape[2]) != self.resize.output_size:
                 x = np.array([self.resize(sample) for sample in x])
-
+            # Data type & 255 division
+            if x.dtype == np.uint8:
+                x = x.astype(np.float32) / 255.0
             batches = [x]
 
         elif isinstance(x, list) and all(isinstance(sample, np.ndarray) for sample in x):