fix: removes multithreading for OCR (#65)

* fix: removes multithreading for OCR

* fix: added missing execution provider for ONNX

* Removed unused dependencies

* fix: removed MockPool as Pool object isn't used anymore

* Version updated

Author: benjats07

CHANGELOG.md

@@ -1,3 +1,7 @@
+## 0.2.9
+
+* Removed multithreading from OCR (DocumentLayout.get_elements_from_layout)
+
 ## 0.2.8
 
 * Refactored YoloX inference code to integrate better with framework

test_unstructured_inference/inference/test_layout.py

@@ -101,7 +101,6 @@ def test_get_page_elements_with_ocr(monkeypatch):
 
     monkeypatch.setattr(detectron2, "is_detectron2_available", lambda *args: True)
     monkeypatch.setattr(layout, "ocr", lambda *args: "An Even Catchier Title")
-    monkeypatch.setattr(layout, "Pool", MockPool)
 
     image = Image.fromarray(np.random.randint(12, 14, size=(40, 10, 3)), mode="RGB")
     print(layout.ocr(text_block, image))

unstructured_inference/__version__.py

@@ -1 +1 @@
-__version__ = "0.2.8"  # pragma: no cover
+__version__ = "0.2.9"  # pragma: no cover

unstructured_inference/inference/layout.py

@@ -1,10 +1,9 @@
 from __future__ import annotations
 from dataclasses import dataclass
-from functools import partial
-from multiprocessing import Pool
 import os
 import re
 import tempfile
+from tqdm import tqdm
 from typing import List, Optional, Tuple, Union, BinaryIO
 
 from layoutparser.io.pdf import load_pdf
@@ -178,18 +177,9 @@ def get_elements_from_layout(self, layout: Layout) -> List[LayoutElement]:
         # sophisticated ordering logic for more complicated layouts.
         layout.sort(key=lambda element: element.coordinates[1], inplace=True)
         # NOTE(benjamin): Creates a Pool for concurrent processing of image elements by OCR
-        pool = Pool()
-        try:
-            get_element_partial = partial(
-                get_element_from_block,
-                image=self.image,
-                layout=self.layout,
-                ocr_strategy=self.ocr_strategy,
-            )
-            elements = pool.map(get_element_partial, layout)
-        finally:
-            pool.close()
-            pool.join()
+        elements = []
+        for e in tqdm(layout):
+            elements.append(get_element_from_block(e, self.image, self.layout, self.ocr_strategy))
         return elements
 
     def _get_image_array(self) -> Union[np.ndarray, None]:

unstructured_inference/models/yolox.py

@@ -51,7 +51,7 @@ def predict(self, x: Image):
         return self.image_processing(x)
 
     def initialize(self, model_path: str, label_map: dict):
-        self.model = onnxruntime.InferenceSession(model_path)
+        self.model = onnxruntime.InferenceSession(model_path, providers=["CPUExecutionProvider"])
         self.layout_classes = label_map
 
     def image_processing(