deepdoctection 0.26__tar.gz → 0.27__tar.gz

{deepdoctection-0.26 → deepdoctection-0.27}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: deepdoctection
-Version: 0.26
+Version: 0.27
 Summary: Repository for Document AI
 Home-page: https://github.com/deepdoctection/deepdoctection
 Author: Dr. Janis Meyer
@@ -13,17 +13,136 @@ Classifier: Programming Language :: Python :: 3.8
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Requires-Python: >=3.8
 Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: catalogue==2.0.7
+Requires-Dist: huggingface_hub>=0.12.0
+Requires-Dist: importlib-metadata>=4.11.2
+Requires-Dist: jsonlines==3.0.0
+Requires-Dist: mock==4.0.3
+Requires-Dist: networkx>=2.7.1
+Requires-Dist: numpy>=1.21
+Requires-Dist: packaging>=20.0
+Requires-Dist: Pillow>=10.0.0
+Requires-Dist: pypdf2
+Requires-Dist: pyyaml==6.0
+Requires-Dist: pyzmq>=16
+Requires-Dist: termcolor>=1.1
+Requires-Dist: tabulate>=0.7.7
+Requires-Dist: tqdm==4.64.0
 Provides-Extra: tf
+Requires-Dist: catalogue==2.0.7; extra == "tf"
+Requires-Dist: huggingface_hub>=0.12.0; extra == "tf"
+Requires-Dist: importlib-metadata>=4.11.2; extra == "tf"
+Requires-Dist: jsonlines==3.0.0; extra == "tf"
+Requires-Dist: mock==4.0.3; extra == "tf"
+Requires-Dist: networkx>=2.7.1; extra == "tf"
+Requires-Dist: numpy>=1.21; extra == "tf"
+Requires-Dist: packaging>=20.0; extra == "tf"
+Requires-Dist: Pillow>=10.0.0; extra == "tf"
+Requires-Dist: pypdf2; extra == "tf"
+Requires-Dist: pyyaml==6.0; extra == "tf"
+Requires-Dist: pyzmq>=16; extra == "tf"
+Requires-Dist: termcolor>=1.1; extra == "tf"
+Requires-Dist: tabulate>=0.7.7; extra == "tf"
+Requires-Dist: tqdm==4.64.0; extra == "tf"
+Requires-Dist: tensorpack; extra == "tf"
+Requires-Dist: protobuf==3.20.1; extra == "tf"
+Requires-Dist: tensorflow-addons>=0.17.1; extra == "tf"
+Requires-Dist: tf2onnx>=1.9.2; extra == "tf"
+Requires-Dist: python-doctr==0.7.0; extra == "tf"
+Requires-Dist: pycocotools>=2.0.2; extra == "tf"
+Requires-Dist: boto3; extra == "tf"
+Requires-Dist: pdfplumber>=0.7.1; extra == "tf"
+Requires-Dist: fasttext; extra == "tf"
+Requires-Dist: jdeskew; extra == "tf"
+Requires-Dist: apted==1.0.3; extra == "tf"
+Requires-Dist: distance==0.1.3; extra == "tf"
+Requires-Dist: lxml>=4.9.1; extra == "tf"
 Provides-Extra: pt
+Requires-Dist: catalogue==2.0.7; extra == "pt"
+Requires-Dist: huggingface_hub>=0.12.0; extra == "pt"
+Requires-Dist: importlib-metadata>=4.11.2; extra == "pt"
+Requires-Dist: jsonlines==3.0.0; extra == "pt"
+Requires-Dist: mock==4.0.3; extra == "pt"
+Requires-Dist: networkx>=2.7.1; extra == "pt"
+Requires-Dist: numpy>=1.21; extra == "pt"
+Requires-Dist: packaging>=20.0; extra == "pt"
+Requires-Dist: Pillow>=10.0.0; extra == "pt"
+Requires-Dist: pypdf2; extra == "pt"
+Requires-Dist: pyyaml==6.0; extra == "pt"
+Requires-Dist: pyzmq>=16; extra == "pt"
+Requires-Dist: termcolor>=1.1; extra == "pt"
+Requires-Dist: tabulate>=0.7.7; extra == "pt"
+Requires-Dist: tqdm==4.64.0; extra == "pt"
+Requires-Dist: timm; extra == "pt"
+Requires-Dist: transformers; extra == "pt"
+Requires-Dist: python-doctr==0.7.0; extra == "pt"
+Requires-Dist: boto3; extra == "pt"
+Requires-Dist: pdfplumber>=0.7.1; extra == "pt"
+Requires-Dist: fasttext; extra == "pt"
+Requires-Dist: jdeskew; extra == "pt"
+Requires-Dist: apted==1.0.3; extra == "pt"
+Requires-Dist: distance==0.1.3; extra == "pt"
+Requires-Dist: lxml>=4.9.1; extra == "pt"
 Provides-Extra: docs
+Requires-Dist: tensorpack; extra == "docs"
+Requires-Dist: boto3; extra == "docs"
+Requires-Dist: transformers; extra == "docs"
+Requires-Dist: pdfplumber>=0.7.1; extra == "docs"
+Requires-Dist: lxml>=4.9.1; extra == "docs"
+Requires-Dist: lxml-stubs; extra == "docs"
+Requires-Dist: jdeskew; extra == "docs"
+Requires-Dist: jinja2==3.0.3; extra == "docs"
+Requires-Dist: mkdocs-material; extra == "docs"
+Requires-Dist: mkdocstrings-python; extra == "docs"
+Requires-Dist: griffe==0.25.0; extra == "docs"
 Provides-Extra: dev
+Requires-Dist: click; extra == "dev"
+Requires-Dist: black==23.7.0; extra == "dev"
+Requires-Dist: isort; extra == "dev"
+Requires-Dist: pylint==2.17.4; extra == "dev"
+Requires-Dist: mypy==1.4.1; extra == "dev"
+Requires-Dist: wandb; extra == "dev"
+Requires-Dist: types-PyYAML; extra == "dev"
+Requires-Dist: types-termcolor==1.1.3; extra == "dev"
+Requires-Dist: types-tabulate; extra == "dev"
+Requires-Dist: types-tqdm; extra == "dev"
+Requires-Dist: lxml-stubs; extra == "dev"
+Requires-Dist: types-Pillow; extra == "dev"
+Requires-Dist: types-urllib3; extra == "dev"
 Provides-Extra: test
+Requires-Dist: pytest; extra == "test"
+Requires-Dist: pytest-cov; extra == "test"
 Provides-Extra: hf
-License-File: LICENSE
+Requires-Dist: catalogue==2.0.7; extra == "hf"
+Requires-Dist: huggingface_hub>=0.12.0; extra == "hf"
+Requires-Dist: importlib-metadata>=4.11.2; extra == "hf"
+Requires-Dist: jsonlines==3.0.0; extra == "hf"
+Requires-Dist: mock==4.0.3; extra == "hf"
+Requires-Dist: networkx>=2.7.1; extra == "hf"
+Requires-Dist: numpy>=1.21; extra == "hf"
+Requires-Dist: packaging>=20.0; extra == "hf"
+Requires-Dist: Pillow>=10.0.0; extra == "hf"
+Requires-Dist: pypdf2; extra == "hf"
+Requires-Dist: pyyaml==6.0; extra == "hf"
+Requires-Dist: pyzmq>=16; extra == "hf"
+Requires-Dist: termcolor>=1.1; extra == "hf"
+Requires-Dist: tabulate>=0.7.7; extra == "hf"
+Requires-Dist: tqdm==4.64.0; extra == "hf"
+Requires-Dist: timm; extra == "hf"
+Requires-Dist: transformers; extra == "hf"
+Requires-Dist: python-doctr==0.7.0; extra == "hf"
+Requires-Dist: boto3; extra == "hf"
+Requires-Dist: pdfplumber>=0.7.1; extra == "hf"
+Requires-Dist: fasttext; extra == "hf"
+Requires-Dist: jdeskew; extra == "hf"
+Requires-Dist: apted==1.0.3; extra == "hf"
+Requires-Dist: distance==0.1.3; extra == "hf"
+Requires-Dist: lxml>=4.9.1; extra == "hf"
 
 
 <p align="center">
-  <img src="https://github.com/deepdoctection/deepdoctection/blob/master/docs/tutorials/_imgs/dd_logo.png" alt="Deep Doctection Logo" width="60%">
+  <img src="https://github.com/deepdoctection/deepdoctection/raw/master/docs/tutorials/_imgs/dd_logo.png" alt="Deep Doctection Logo" width="60%">
   <h3 align="center">
   A Document AI Package
   </h3>
@@ -123,20 +242,20 @@ plt.axis('off')
 plt.imshow(image)
 ```
 
-![text](./docs/tutorials/_imgs/dd_rm_sample.png)
+![text](https://github.com/deepdoctection/deepdoctection/raw/master/docs/tutorials/_imgs/dd_rm_sample.png)
 
 ```
 HTML(page.tables[0].html)
 ```
 
-![table](./docs/tutorials/_imgs/dd_rm_table.png)
+![table](https://github.com/deepdoctection/deepdoctection/raw/master/docs/tutorials/_imgs/dd_rm_table.png)
 
 
 ```
 print(page.text)
 ```
 
-![table](./docs/tutorials/_imgs/dd_rm_text.png)
+![table](https://github.com/deepdoctection/deepdoctection/raw/master/docs/tutorials/_imgs/dd_rm_text.png)
  
 
 ## Documentation
@@ -150,14 +269,14 @@ documentation.
 
 ## Requirements
 
-![requirements](./docs/tutorials/_imgs/requirements_deepdoctection.png)
+![requirements](https://github.com/deepdoctection/deepdoctection/raw/master/docs/tutorials/_imgs/requirements_deepdoctection.png)
 
 Everything in the overview listed below the **deep**doctection layer are necessary requirements and have to be installed 
 separately. 
 
 - Linux or macOS. (Windows is not supported but there is a [Dockerfile](./docker/pytorch-cpu-jupyter/Dockerfile) available)
 - Python >= 3.8
-- 1.8 <= PyTorch < 2.0 **or** Tensorflow >= 2.9 and CUDA. If you want to run the models provided by Tensorpack a GPU is
+- 1.12 <= PyTorch < 2.0 **or** Tensorflow >= 2.9 and CUDA. If you want to run the models provided by Tensorpack a GPU is
   required. You can run on PyTorch with a CPU only.
 - **deep**doctection uses Python wrappers for [Poppler](https://poppler.freedesktop.org/) to convert PDF documents into 
 images. 
@@ -230,7 +349,7 @@ pip install deepdoctection
 This will ignore all model libraries (layers above the **deep**doctection layer in the diagram) and you 
 will be responsible to install them by yourself. Note, that you will not be able to run any pipeline with this setup.
 
-For further information, please consult the [**full installation instructions**](https://deepdoctection.readthedocs.io/en/latest/manual/install.html).
+For further information, please consult the [**full installation instructions**](https://deepdoctection.readthedocs.io/en/latest/install/).
 
 
 ### Installation from source

{deepdoctection-0.26 → deepdoctection-0.27}/README.md

@@ -1,6 +1,6 @@
 
 <p align="center">
-  <img src="https://github.com/deepdoctection/deepdoctection/blob/master/docs/tutorials/_imgs/dd_logo.png" alt="Deep Doctection Logo" width="60%">
+  <img src="https://github.com/deepdoctection/deepdoctection/raw/master/docs/tutorials/_imgs/dd_logo.png" alt="Deep Doctection Logo" width="60%">
   <h3 align="center">
   A Document AI Package
   </h3>
@@ -100,20 +100,20 @@ plt.axis('off')
 plt.imshow(image)
 ```
 
-![text](./docs/tutorials/_imgs/dd_rm_sample.png)
+![text](https://github.com/deepdoctection/deepdoctection/raw/master/docs/tutorials/_imgs/dd_rm_sample.png)
 
 ```
 HTML(page.tables[0].html)
 ```
 
-![table](./docs/tutorials/_imgs/dd_rm_table.png)
+![table](https://github.com/deepdoctection/deepdoctection/raw/master/docs/tutorials/_imgs/dd_rm_table.png)
 
 
 ```
 print(page.text)
 ```
 
-![table](./docs/tutorials/_imgs/dd_rm_text.png)
+![table](https://github.com/deepdoctection/deepdoctection/raw/master/docs/tutorials/_imgs/dd_rm_text.png)
  
 
 ## Documentation
@@ -127,14 +127,14 @@ documentation.
 
 ## Requirements
 
-![requirements](./docs/tutorials/_imgs/requirements_deepdoctection.png)
+![requirements](https://github.com/deepdoctection/deepdoctection/raw/master/docs/tutorials/_imgs/requirements_deepdoctection.png)
 
 Everything in the overview listed below the **deep**doctection layer are necessary requirements and have to be installed 
 separately. 
 
 - Linux or macOS. (Windows is not supported but there is a [Dockerfile](./docker/pytorch-cpu-jupyter/Dockerfile) available)
 - Python >= 3.8
-- 1.8 <= PyTorch < 2.0 **or** Tensorflow >= 2.9 and CUDA. If you want to run the models provided by Tensorpack a GPU is
+- 1.12 <= PyTorch < 2.0 **or** Tensorflow >= 2.9 and CUDA. If you want to run the models provided by Tensorpack a GPU is
   required. You can run on PyTorch with a CPU only.
 - **deep**doctection uses Python wrappers for [Poppler](https://poppler.freedesktop.org/) to convert PDF documents into 
 images. 
@@ -207,7 +207,7 @@ pip install deepdoctection
 This will ignore all model libraries (layers above the **deep**doctection layer in the diagram) and you 
 will be responsible to install them by yourself. Note, that you will not be able to run any pipeline with this setup.
 
-For further information, please consult the [**full installation instructions**](https://deepdoctection.readthedocs.io/en/latest/manual/install.html).
+For further information, please consult the [**full installation instructions**](https://deepdoctection.readthedocs.io/en/latest/install/).
 
 
 ### Installation from source

{deepdoctection-0.26 → deepdoctection-0.27}/deepdoctection/__init__.py

@@ -14,7 +14,7 @@ from packaging import version
 from .utils.file_utils import _LazyModule, get_tf_version, pytorch_available, tf_available
 from .utils.logger import logger
 
-__version__ = 0.26
+__version__ = 0.27
 
 _IMPORT_STRUCTURE = {
     "analyzer": ["get_dd_analyzer", "build_analyzer"],
@@ -311,6 +311,10 @@ _IMPORT_STRUCTURE = {
         "get_fasttext_requirement",
         "wandb_available",
         "get_wandb_requirement",
+        "opencv_available",
+        "get_opencv_requirement",
+        "pillow_available",
+        "get_pillow_requirement",
         "load_image_from_file",
         "load_bytes_from_pdf_file",
         "get_load_image_func",
@@ -378,6 +382,7 @@ _IMPORT_STRUCTURE = {
         "draw_text",
         "draw_boxes",
         "interactive_imshow",
+        "viz_handler",
     ],
 }
 
@@ -403,6 +408,7 @@ if tf_available():
         except Exception:  # pylint: disable=W0703
             pass
 
+
 # Direct imports for type-checking
 if TYPE_CHECKING:
     from .analyzer import *

{deepdoctection-0.26 → deepdoctection-0.27}/deepdoctection/analyzer/dd.py

@@ -36,7 +36,7 @@ from ..extern.tessocr import TesseractOcrDetector
 from ..extern.texocr import TextractOcrDetector
 from ..pipe.base import PipelineComponent
 from ..pipe.cell import DetectResultGenerator, SubImageLayoutService
-from ..pipe.common import MatchingService, PageParsingService
+from ..pipe.common import AnnotationNmsService, MatchingService, PageParsingService
 from ..pipe.doctectionpipe import DoctectionPipe
 from ..pipe.layout import ImageLayoutService
 from ..pipe.order import TextOrderService
@@ -206,7 +206,7 @@ def _build_ocr(cfg: AttrDict) -> Union[TesseractOcrDetector, DoctrTextRecognizer
         profile = ModelCatalog.get_profile(weights)
         if profile.architecture is None:
             raise ValueError("model profile.architecture must be specified")
-        return DoctrTextRecognizer(profile.architecture, weights_path, cfg.DEVICE)
+        return DoctrTextRecognizer(profile.architecture, weights_path, cfg.DEVICE, lib=cfg.LIB)
     if cfg.OCR.USE_TEXTRACT:
         credentials_kwargs = {
             "aws_access_key_id": environ.get("ACCESS_KEY"),
@@ -225,7 +225,7 @@ def _build_doctr_word(cfg: AttrDict) -> DoctrTextlineDetector:
         raise ValueError("model profile.architecture must be specified")
     if profile.categories is None:
         raise ValueError("model profile.categories must be specified")
-    return DoctrTextlineDetector(profile.architecture, weights_path, profile.categories, cfg.DEVICE)
+    return DoctrTextlineDetector(profile.architecture, weights_path, profile.categories, cfg.DEVICE, lib=cfg.LIB)
 
 
 def build_analyzer(cfg: AttrDict) -> DoctectionPipe:
@@ -242,6 +242,17 @@ def build_analyzer(cfg: AttrDict) -> DoctectionPipe:
         layout = _build_service(d_layout, cfg, "LAYOUT")
         pipe_component_list.append(layout)
 
+    # setup layout nms service
+    if cfg.LAYOUT_NMS_PAIRS.COMBINATIONS and cfg.USE_LAYOUT:
+        if not isinstance(cfg.LAYOUT_NMS_PAIRS.COMBINATIONS, list) and not isinstance(
+            cfg.LAYOUT_NMS_PAIRS.COMBINATIONS[0], list
+        ):
+            raise ValueError("LAYOUT_NMS_PAIRS mus be a list of lists")
+        layout_nms_serivce = AnnotationNmsService(
+            cfg.LAYOUT_NMS_PAIRS.COMBINATIONS, cfg.LAYOUT_NMS_PAIRS.THRESHOLDS, cfg.LAYOUT_NMS_PAIRS.PRIORITY
+        )
+        pipe_component_list.append(layout_nms_serivce)
+
     # setup tables service
     if cfg.USE_TABLE_SEGMENTATION:
         d_item = _build_detector(cfg, "ITEM")
@@ -302,6 +313,7 @@ def build_analyzer(cfg: AttrDict) -> DoctectionPipe:
         )
         pipe_component_list.append(text)
 
+    if cfg.USE_PDF_MINER or cfg.USE_OCR:
         match = MatchingService(
             parent_categories=cfg.WORD_MATCHING.PARENTAL_CATEGORIES,
             child_categories=LayoutType.word,

{deepdoctection-0.26 → deepdoctection-0.27}/deepdoctection/configs/conf_dd_one.yaml

@@ -33,6 +33,10 @@ PT:
       WEIGHTS: cell/d2_model_1849999_cell_inf_only.pt
       WEIGHTS_TS: cell/d2_model_1849999_cell_inf_only.ts
       FILTER:
+LAYOUT_NMS_PAIRS:
+  COMBINATIONS:
+  THRESHOLDS:
+  PRIORITY:
 SEGMENTATION:
   ASSIGNMENT_RULE: ioa
   THRESHOLD_ROWS: 0.4

{deepdoctection-0.26 → deepdoctection-0.27}/deepdoctection/datapoint/convert.py

@@ -25,7 +25,6 @@ from io import BytesIO
 from shutil import which
 from typing import Any, Optional, Union, no_type_check
 
-import cv2
 import numpy as np
 from numpy import uint8
 from numpy.typing import NDArray
@@ -34,6 +33,7 @@ from PyPDF2 import PdfReader
 from ..utils.detection_types import ImageType
 from ..utils.develop import deprecated
 from ..utils.pdf_utils import pdf_to_np_array
+from ..utils.viz import viz_handler
 
 __all__ = [
     "convert_b64_to_np_array",
@@ -81,9 +81,8 @@ def convert_b64_to_np_array(image: str) -> ImageType:
     :param image: An image as base64 string.
     :return: numpy array.
     """
-    np_array = np.fromstring(base64.b64decode(image), np.uint8)  # type: ignore
-    np_array = cv2.imdecode(np_array, cv2.IMREAD_COLOR).astype(np.float32)
-    return np_array.astype(uint8)
+
+    return viz_handler.convert_b64_to_np(image).astype(uint8)
 
 
 def convert_np_array_to_b64(np_image: ImageType) -> str:
@@ -93,9 +92,7 @@ def convert_np_array_to_b64(np_image: ImageType) -> str:
     :param np_image: An image as numpy array.
     :return: An image as base64 string.
     """
-    np_encode = cv2.imencode(".png", np_image)
-    image = base64.b64encode(np_encode[1]).decode("utf-8")  # type: ignore
-    return image
+    return viz_handler.convert_np_to_b64(np_image)
 
 
 @no_type_check
@@ -106,9 +103,7 @@ def convert_np_array_to_b64_b(np_image: ImageType) -> bytes:
     :param np_image: An image as numpy array.
     :return: An image as base64 bytes.
     """
-    np_encode = cv2.imencode(".png", np_image)
-    b_image = np_encode[1].tobytes()
-    return b_image
+    return viz_handler.encode(np_image)
 
 
 @deprecated("Use convert_pdf_bytes_to_np_array_v2", "2022-02-23")

{deepdoctection-0.26 → deepdoctection-0.27}/deepdoctection/datapoint/image.py

@@ -626,8 +626,8 @@ class Image:
             self.remove_image_from_lower_hierachy()
         export_dict = self.as_dict()
         export_dict["location"] = str(export_dict["location"])
-        if image_to_json and self.image is not None:
-            export_dict["_image"] = convert_np_array_to_b64(self.image)
+        if not image_to_json:
+            export_dict["_image"] = None
         if dry:
             return export_dict
         with open(path_json, "w", encoding="UTF-8") as file:

{deepdoctection-0.26 → deepdoctection-0.27}/deepdoctection/datapoint/view.py

@@ -23,7 +23,6 @@ simplify consumption
 from copy import copy
 from typing import Any, Dict, List, Mapping, Optional, Sequence, Set, Tuple, Type, Union, no_type_check
 
-import cv2
 import numpy as np
 
 from ..utils.detection_types import ImageType, JsonDict, Pathlike
@@ -39,7 +38,7 @@ from ..utils.settings import (
     WordType,
     get_type,
 )
-from ..utils.viz import draw_boxes, interactive_imshow
+from ..utils.viz import draw_boxes, interactive_imshow, viz_handler
 from .annotation import ContainerAnnotation, ImageAnnotation, SummaryAnnotation, ann_from_dict
 from .box import BoundingBox
 from .image import Image
@@ -415,6 +414,17 @@ class Page(Image):
     text_container: ObjectTypes
     floating_text_block_categories: List[ObjectTypes]
     image_orig: Image
+    _attribute_names: Set[str] = {
+        "text",
+        "chunks",
+        "tables",
+        "layouts",
+        "words",
+        "file_name",
+        "location",
+        "document_id",
+        "page_number",
+    }
 
     @no_type_check
     def get_annotation(
@@ -734,7 +744,9 @@ class Page(Image):
                     )
                 else:
                     img = draw_boxes(self.image, boxes, category_names_list)
-                img = cv2.resize(img, None, fx=1.3, fy=1.3, interpolation=cv2.INTER_CUBIC)
+                scale_fx, scale_fy = 1.3, 1.3
+                scaled_width, scaled_height = int(self.width * scale_fx), int(self.height * scale_fy)
+                img = viz_handler.resize(img, scaled_width, scaled_height, "VIZ")
             else:
                 img = self.image
 
@@ -744,24 +756,32 @@ class Page(Image):
             return img
         return None
 
-    @staticmethod
-    def get_attribute_names() -> Set[str]:
+    @classmethod
+    def get_attribute_names(cls) -> Set[str]:
         """
         :return: A set of registered attributes.
         """
-        return set(PageType).union(
-            {
-                "text",
-                "chunks",
-                "tables",
-                "layouts",
-                "words",
-                "file_name",
-                "location",
-                "document_id",
-                "page_number",
-            }
-        )
+        return set(PageType).union(cls._attribute_names)
+
+    @classmethod
+    def add_attribute_name(cls, attribute_name: Union[str, ObjectTypes]) -> None:
+        """
+        Adding a custom attribute name to a Page class.
+
+                **Example:**
+
+                Page.add_attribute_name("foo")
+
+                page = Page.from_image(...)
+                print(page.foo)
+
+        Note, that the attribute must be registered as a valid `ObjectTypes`
+
+        :param attribute_name: attribute name to add
+        """
+
+        attribute_name = get_type(attribute_name)
+        cls._attribute_names.add(attribute_name.value)
 
     def save(
         self,

{deepdoctection-0.26 → deepdoctection-0.27}/deepdoctection/datasets/save.py

@@ -23,13 +23,12 @@ import json
 from pathlib import Path
 from typing import Optional
 
-from cv2 import imwrite
-
 from ..dataflow import DataFlow, MapData, SerializerJsonlines
 from ..datapoint.convert import convert_b64_to_np_array
 from ..datapoint.image import Image
 from ..utils.detection_types import JsonDict, Pathlike
 from ..utils.fs import mkdir_p
+from ..utils.viz import viz_handler
 
 
 def dataflow_to_json(
@@ -84,7 +83,8 @@ def dataflow_to_json(
                 target_file_png = path / "image" / (dp["file_name"].split(".")[0] + ".png")
                 image = dp.pop("_image")
                 image = convert_b64_to_np_array(image)
-                imwrite(str(target_file_png), image)
+
+                viz_handler.write_image(str(target_file_png), image)
 
             with open(target_file, "w", encoding="UTF-8") as file:
                 json.dump(dp, file)

{deepdoctection-0.26 → deepdoctection-0.27}/deepdoctection/extern/d2detect.py

@@ -23,7 +23,6 @@ from copy import copy
 from pathlib import Path
 from typing import Any, Dict, List, Literal, Mapping, Optional, Sequence
 
-import cv2
 import numpy as np
 
 from ..utils.detection_types import ImageType, Requirement
@@ -130,7 +129,7 @@ def d2_jit_predict_image(
         keep = batched_nms(boxes, scores, class_masks, nms_thresh_class_agnostic).cpu()
 
         # The exported model does not contain the final resize step, so we need to add it manually here
-        inverse_resizer = ResizeTransform(new_height, new_width, height, width, cv2.INTER_LINEAR)
+        inverse_resizer = ResizeTransform(new_height, new_width, height, width, "VIZ")
         np_boxes = np.reshape(boxes.cpu().numpy(), (-1, 2))
         np_boxes = inverse_resizer.apply_coords(np_boxes)
         np_boxes = np.reshape(np_boxes, (-1, 4))

{deepdoctection-0.26 → deepdoctection-0.27}/deepdoctection/extern/doctrocr.py

@@ -62,14 +62,14 @@ def _set_device_str(device: Optional[str] = None) -> str:
     return device
 
 
-def _load_model(path_weights: str, doctr_predictor: Any, device: str) -> None:
-    if pytorch_available():
+def _load_model(path_weights: str, doctr_predictor: Any, device: str, lib: str) -> None:
+    if lib == "PT" and pytorch_available():
         state_dict = torch.load(path_weights, map_location=device)
         for key in list(state_dict.keys()):
             state_dict["model." + key] = state_dict.pop(key)
         doctr_predictor.load_state_dict(state_dict)
         doctr_predictor.to(device)
-    elif tf_available():
+    elif lib == "TF" and tf_available():
         # Unzip the archive
         params_path = Path(path_weights).parent
         is_zip_path = path_weights.endswith(".zip")
@@ -99,7 +99,7 @@ def doctr_predict_text_lines(np_img: ImageType, predictor: "DetectionPredictor",
         DetectionResult(
             box=box[:4].tolist(), class_id=1, score=box[4], absolute_coords=False, class_name=LayoutType.word
         )
-        for box in raw_output[0]
+        for box in raw_output[0]["words"]
     ]
     return detection_results
 
@@ -173,7 +173,9 @@ class DoctrTextlineDetector(ObjectDetector):
         path_weights: str,
         categories: Mapping[str, TypeOrStr],
         device: Optional[Literal["cpu", "cuda"]] = None,
+        lib: str = "TF",
     ) -> None:
+        self.lib = lib
         self.name = "doctr_text_detector"
         self.architecture = architecture
         self.path_weights = path_weights
@@ -205,14 +207,14 @@ class DoctrTextlineDetector(ObjectDetector):
         raise ModuleNotFoundError("Neither Tensorflow nor PyTorch has been installed. Cannot use DoctrTextlineDetector")
 
     def clone(self) -> PredictorBase:
-        return self.__class__(self.architecture, self.path_weights, self.categories, self.device_input)
+        return self.__class__(self.architecture, self.path_weights, self.categories, self.device_input, self.lib)
 
     def possible_categories(self) -> List[ObjectTypes]:
         return [LayoutType.word]
 
     def load_model(self) -> None:
         """Loading model weights"""
-        _load_model(self.path_weights, self.doctr_predictor, self.device)
+        _load_model(self.path_weights, self.doctr_predictor, self.device, self.lib)
 
 
 class DoctrTextRecognizer(TextRecognizer):
@@ -252,7 +254,10 @@ class DoctrTextRecognizer(TextRecognizer):
 
     """
 
-    def __init__(self, architecture: str, path_weights: str, device: Optional[Literal["cpu", "cuda"]] = None) -> None:
+    def __init__(
+        self, architecture: str, path_weights: str, device: Optional[Literal["cpu", "cuda"]] = None, lib: str = "TF"
+    ) -> None:
+        self.lib = lib
         self.name = "doctr_text_recognizer"
         self.architecture = architecture
         self.path_weights = path_weights
@@ -281,8 +286,8 @@ class DoctrTextRecognizer(TextRecognizer):
         raise ModuleNotFoundError("Neither Tensorflow nor PyTorch has been installed. Cannot use DoctrTextRecognizer")
 
     def clone(self) -> PredictorBase:
-        return self.__class__(self.architecture, self.path_weights, self.device_input)
+        return self.__class__(self.architecture, self.path_weights, self.device_input, self.lib)
 
     def load_model(self) -> None:
         """Loading model weights"""
-        _load_model(self.path_weights, self.doctr_predictor, self.device)
+        _load_model(self.path_weights, self.doctr_predictor, self.device, self.lib)

{deepdoctection-0.26 → deepdoctection-0.27}/deepdoctection/extern/tp/tpfrcnn/common.py

@@ -10,7 +10,6 @@ This file is modified from
 """
 
 
-import cv2
 import numpy as np
 from tensorpack.dataflow.imgaug import ImageAugmentor, ResizeTransform  # pylint: disable=E0401
 
@@ -25,11 +24,11 @@ class CustomResize(ImageAugmentor):
     Try resizing the shortest edge to a certain number while avoiding the longest edge to exceed max_size.
     """
 
-    def __init__(self, short_edge_length, max_size, interp=cv2.INTER_LINEAR):
+    def __init__(self, short_edge_length, max_size, interp=1):
         """
         :param short_edge_length: a [min, max] interval from which to sample the shortest edge length.
         :param max_size: maximum allowed longest edge length.
-        :param interp: cv2 interpolation mode
+        :param interp: Interpolation mode. We use Tensorpack's internal `ResizeTransform`, that always requires OpenCV
         """
         super().__init__()
         if isinstance(short_edge_length, int):