fix: typeerror when using chipper (#311)

This PR resolves #310 
- chipper, or any page layout extracted with an element extraction model
do not have key attributes like `image_metadata` populated
- this leads to `None` values for image width and height, which lead to
the bug
- this fix prevents the function early return after chipper finds the
elements
- it continues the logic to allow other key attributes of the page to be
filled
- a bonus from this fix is we remove the image data from the page (which
is not needed downstream) for chipper generated pages

## Test

A unit test is modified to test all the routes, including using an
element extraction model, for page layout

Additionally grab this attached pdf and when running partition using
chipper the main branch would lead to type error but this fix would run
without error.

[005-CISA-AA22-076-Strengthening-Cybersecurity-p1-p4.pdf](https://github.com/Unstructured-IO/unstructured-inference/files/13731533/005-CISA-AA22-076-Strengthening-Cybersecurity-p1-p4.pdf)

Author: badGarnet

CHANGELOG.md

@@ -1,3 +1,7 @@
+## 0.7.21
+
+* fix: fix a bug where chipper, or any element extraction model based `PageLayout` object, lack `image_metadata` and other attributes that are required for downstream processing; this fix also reduces the memory overhead of using chipper model
+
 ## 0.7.20
 
 * chipper-v3: improved table prediction

test_unstructured_inference/inference/test_layout.py

@@ -9,7 +9,10 @@
 
 import unstructured_inference.models.base as models
 from unstructured_inference.inference import elements, layout, layoutelement
-from unstructured_inference.inference.elements import EmbeddedTextRegion, ImageTextRegion
+from unstructured_inference.inference.elements import (
+    EmbeddedTextRegion,
+    ImageTextRegion,
+)
 from unstructured_inference.models.unstructuredmodel import (
     UnstructuredElementExtractionModel,
     UnstructuredObjectDetectionModel,
@@ -271,12 +274,14 @@ def filter_by(self, *args, **kwargs):
         return MockLayout()
 
 
+@pytest.mark.parametrize("element_extraction_model", [None, "foo"])
 @pytest.mark.parametrize("filetype", ["png", "jpg", "tiff"])
-def test_from_image_file(monkeypatch, mock_final_layout, filetype):
+def test_from_image_file(monkeypatch, mock_final_layout, filetype, element_extraction_model):
     def mock_get_elements(self, *args, **kwargs):
         self.elements = [mock_final_layout]
 
     monkeypatch.setattr(layout.PageLayout, "get_elements_with_detection_model", mock_get_elements)
+    monkeypatch.setattr(layout.PageLayout, "get_elements_using_image_extraction", mock_get_elements)
     filename = f"sample-docs/loremipsum.{filetype}"
     image = Image.open(filename)
     image_metadata = {
@@ -285,7 +290,10 @@ def mock_get_elements(self, *args, **kwargs):
         "height": image.height,
     }
 
-    doc = layout.DocumentLayout.from_image_file(filename)
+    doc = layout.DocumentLayout.from_image_file(
+        filename,
+        element_extraction_model=element_extraction_model,
+    )
     page = doc.pages[0]
     assert page.elements[0] == mock_final_layout
     assert page.image is None

unstructured_inference/__version__.py

@@ -1 +1 @@
-__version__ = "0.7.20"  # pragma: no cover
+__version__ = "0.7.21"  # pragma: no cover

unstructured_inference/inference/layout.py

@@ -322,10 +322,10 @@ def from_image(
             detection_model=detection_model,
             element_extraction_model=element_extraction_model,
         )
+        # FIXME (yao): refactor the other methods so they all return elements like the third route
         if page.element_extraction_model is not None:
             page.get_elements_using_image_extraction()
-            return page
-        if fixed_layout is None:
+        elif fixed_layout is None:
             page.get_elements_with_detection_model()
         else:
             page.elements = page.get_elements_from_layout(fixed_layout)