Robust pdf loading for empty pages (#115)

* more robust PDF loading for "empty PDFs"

* Tests

* fix tests

Author: lolipopshock

src/layoutparser/io/pdf.py

@@ -53,6 +53,10 @@ def extract_words_for_page(
     )
 
     df = pd.DataFrame(tokens)
+    
+    if len(df) == 0:
+        return Layout()
+    
     df[["x0", "x1"]] = (
         df[["x0", "x1"]].clip(lower=0, upper=int(page.width)).astype("float")
     )

tests/fixtures/io/empty.pdf

@@ -78,4 +78,11 @@ def test_pdf():
         assert attr_name in page_layout.page_data
 
     assert len(set(ele.type for ele in page_layout)) == 3
-    # Only three types of font show-up in the file
+    # Only three types of font show-up in the file
+    
+def test_empty_pdf():
+    pdf_layout = load_pdf("tests/fixtures/io/empty.pdf")
+    assert len(pdf_layout) == 1 # Only one page
+    
+    page_layout = pdf_layout[0]
+    assert len(page_layout) == 0 # No selectable tokens on the page