Support bytes and file objects in load_pdf

Allows load_pdf to accept raw bytes or binary file-like objects in
addition to file paths, useful for processing cloud-stored PDFs.

Based on: Layout-Parser#153 (davibarreira)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: IgorTavcar

src/layoutparser/io/pdf.py

@@ -12,7 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import List, Union, Optional, Dict, Tuple
+from io import BytesIO
+from typing import BinaryIO, List, Union, Optional, Dict, Tuple
 
 import pdfplumber
 import pandas as pd
@@ -84,7 +85,7 @@ def extract_words_for_page(
 
 
 def load_pdf(
-    filename: str,
+    filename: Union[str, bytes, BinaryIO],
     load_images: bool = False,
     x_tolerance: int = 1.5,
     y_tolerance: int = 2,
@@ -99,7 +100,8 @@ def load_pdf(
     in a list of Layout objects with the original page order.
 
     Args:
-        filename (str): The path to the PDF file.
+        filename (Union[str, bytes, BinaryIO]): The path to the PDF file,
+            raw bytes, or a binary file-like object.
         load_images (bool, optional):
             Whether load screenshot for each page of the PDF file.
             When set to true, the function will return both the layout and
@@ -180,6 +182,10 @@ def load_pdf(
         >>> lp.draw_box(pdf_images[0], pdf_layout[0])
     """
 
+    is_path = isinstance(filename, str)
+    if isinstance(filename, bytes):
+        filename = BytesIO(filename)
+
     plumber_pdf_object = pdfplumber.open(filename)
 
     all_page_layout = []
@@ -211,7 +217,12 @@ def load_pdf(
     else:
         import pdf2image
 
-        pdf_images = pdf2image.convert_from_path(filename, dpi=dpi)
+        if is_path:
+            pdf_images = pdf2image.convert_from_path(filename, dpi=dpi)
+        else:
+            if isinstance(filename, BytesIO):
+                filename.seek(0)
+            pdf_images = pdf2image.convert_from_bytes(filename.read() if hasattr(filename, 'read') else filename, dpi=dpi)
 
         for page_id, page_image in enumerate(pdf_images):
             image_width, image_height = page_image.size

tests/test_io.py

@@ -103,4 +103,29 @@ def test_pdf_closes_file_handle():
         load_pdf("dummy.pdf")
 
         mock_pdf.__enter__.assert_called_once()
-        mock_pdf.__exit__.assert_called_once()
+        mock_pdf.__exit__.assert_called_once()
+
+
+def test_pdf_with_bytes():
+    with open("tests/fixtures/io/example.pdf", "rb") as f:
+        pdf_bytes = f.read()
+
+    pdf_layout = load_pdf(pdf_bytes)
+    assert len(pdf_layout) == 1
+
+    page_layout = pdf_layout[0]
+    for attr_name in ["width", "height", "index"]:
+        assert attr_name in page_layout.page_data
+
+    assert len(set(ele.type for ele in page_layout)) == 3
+
+
+def test_pdf_with_file_object():
+    with open("tests/fixtures/io/example.pdf", "rb") as f:
+        pdf_layout = load_pdf(f)
+
+    assert len(pdf_layout) == 1
+
+    page_layout = pdf_layout[0]
+    for attr_name in ["width", "height", "index"]:
+        assert attr_name in page_layout.page_data