Vik Paruchuri
commited on
Commit
·
906ae67
1
Parent(s):
2b12856
Add debug mode to marker app
Browse files- marker/processors/code.py +1 -1
- marker/processors/debug.py +2 -0
- marker/processors/footnote.py +4 -4
- marker/renderers/__init__.py +6 -2
- marker/schema/document.py +1 -0
- marker/settings.py +1 -0
- marker_app.py +25 -2
marker/processors/code.py
CHANGED
|
@@ -47,4 +47,4 @@ class CodeProcessor(BaseProcessor):
|
|
| 47 |
code_text += text
|
| 48 |
is_new_line = text.endswith("\n")
|
| 49 |
|
| 50 |
-
block.code = code_text
|
|
|
|
| 47 |
code_text += text
|
| 48 |
is_new_line = text.endswith("\n")
|
| 49 |
|
| 50 |
+
block.code = code_text.rstrip()
|
marker/processors/debug.py
CHANGED
|
@@ -54,6 +54,8 @@ class DebugProcessor(BaseProcessor):
|
|
| 54 |
if any([self.debug_layout_images, self.debug_pdf_images, self.debug_json]):
|
| 55 |
os.makedirs(self.debug_folder, exist_ok=True)
|
| 56 |
|
|
|
|
|
|
|
| 57 |
if self.debug_layout_images:
|
| 58 |
self.draw_layout_debug_images(document)
|
| 59 |
print(f"Dumped layout debug images to {self.debug_data_folder}")
|
|
|
|
| 54 |
if any([self.debug_layout_images, self.debug_pdf_images, self.debug_json]):
|
| 55 |
os.makedirs(self.debug_folder, exist_ok=True)
|
| 56 |
|
| 57 |
+
document.debug_data_path = self.debug_folder
|
| 58 |
+
|
| 59 |
if self.debug_layout_images:
|
| 60 |
self.draw_layout_debug_images(document)
|
| 61 |
print(f"Dumped layout debug images to {self.debug_data_folder}")
|
marker/processors/footnote.py
CHANGED
|
@@ -19,15 +19,15 @@ class FootnoteProcessor(BaseProcessor):
|
|
| 19 |
Attributes:
|
| 20 |
page_bottom_threshold (float):
|
| 21 |
The fraction of page height that is considered the bottom.
|
| 22 |
-
Default is .
|
| 23 |
|
| 24 |
line_height_scaler (float):
|
| 25 |
-
The amount to scale line height by to consider a block a footnote.
|
| 26 |
-
Default is .
|
| 27 |
"""
|
| 28 |
block_types = (BlockTypes.Footnote,)
|
| 29 |
page_bottom_threshold = .75
|
| 30 |
-
line_height_scaler = .
|
| 31 |
|
| 32 |
|
| 33 |
def __call__(self, document: Document):
|
|
|
|
| 19 |
Attributes:
|
| 20 |
page_bottom_threshold (float):
|
| 21 |
The fraction of page height that is considered the bottom.
|
| 22 |
+
Default is .8
|
| 23 |
|
| 24 |
line_height_scaler (float):
|
| 25 |
+
The amount to scale line height by to consider a block a footnote. (from N to 1+(1-N))
|
| 26 |
+
Default is .99
|
| 27 |
"""
|
| 28 |
block_types = (BlockTypes.Footnote,)
|
| 29 |
page_bottom_threshold = .75
|
| 30 |
+
line_height_scaler = .99
|
| 31 |
|
| 32 |
|
| 33 |
def __call__(self, document: Document):
|
marker/renderers/__init__.py
CHANGED
|
@@ -63,10 +63,14 @@ class BaseRenderer:
|
|
| 63 |
return page_stats
|
| 64 |
|
| 65 |
def generate_document_metadata(self, document, document_output):
|
| 66 |
-
|
| 67 |
"table_of_contents": document.table_of_contents,
|
| 68 |
-
"page_stats": self.generate_page_stats(document, document_output)
|
| 69 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
|
| 71 |
def extract_block_html(self, document, block_output):
|
| 72 |
soup = BeautifulSoup(block_output.html, 'html.parser')
|
|
|
|
| 63 |
return page_stats
|
| 64 |
|
| 65 |
def generate_document_metadata(self, document, document_output):
|
| 66 |
+
metadata = {
|
| 67 |
"table_of_contents": document.table_of_contents,
|
| 68 |
+
"page_stats": self.generate_page_stats(document, document_output),
|
| 69 |
}
|
| 70 |
+
if document.debug_data_path is not None:
|
| 71 |
+
metadata["debug_data_path"] = document.debug_data_path
|
| 72 |
+
|
| 73 |
+
return metadata
|
| 74 |
|
| 75 |
def extract_block_html(self, document, block_output):
|
| 76 |
soup = BeautifulSoup(block_output.html, 'html.parser')
|
marker/schema/document.py
CHANGED
|
@@ -27,6 +27,7 @@ class Document(BaseModel):
|
|
| 27 |
pages: List[PageGroup]
|
| 28 |
block_type: BlockTypes = BlockTypes.Document
|
| 29 |
table_of_contents: List[TocItem] | None = None
|
|
|
|
| 30 |
|
| 31 |
def get_block(self, block_id: BlockId):
|
| 32 |
page = self.get_page(block_id.page_id)
|
|
|
|
| 27 |
pages: List[PageGroup]
|
| 28 |
block_type: BlockTypes = BlockTypes.Document
|
| 29 |
table_of_contents: List[TocItem] | None = None
|
| 30 |
+
debug_data_path: str | None = None # Path that debug data was saved to
|
| 31 |
|
| 32 |
def get_block(self, block_id: BlockId):
|
| 33 |
page = self.get_page(block_id.page_id)
|
marker/settings.py
CHANGED
|
@@ -12,6 +12,7 @@ class Settings(BaseSettings):
|
|
| 12 |
BASE_DIR: str = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
|
| 13 |
OUTPUT_DIR: str = os.path.join(BASE_DIR, "conversion_results")
|
| 14 |
FONT_DIR: str = os.path.join(BASE_DIR, "static", "fonts")
|
|
|
|
| 15 |
|
| 16 |
# General models
|
| 17 |
TORCH_DEVICE: Optional[str] = None # Note: MPS device does not work for text detection, and will default to CPU
|
|
|
|
| 12 |
BASE_DIR: str = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
|
| 13 |
OUTPUT_DIR: str = os.path.join(BASE_DIR, "conversion_results")
|
| 14 |
FONT_DIR: str = os.path.join(BASE_DIR, "static", "fonts")
|
| 15 |
+
DEBUG_DATA_FOLDER: str = os.path.join(BASE_DIR, "debug_data")
|
| 16 |
|
| 17 |
# General models
|
| 18 |
TORCH_DEVICE: Optional[str] = None # Note: MPS device does not work for text detection, and will default to CPU
|
marker_app.py
CHANGED
|
@@ -1,5 +1,7 @@
|
|
| 1 |
import os
|
| 2 |
|
|
|
|
|
|
|
| 3 |
os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"
|
| 4 |
os.environ["IN_STREAMLIT"] = "true"
|
| 5 |
|
|
@@ -11,6 +13,7 @@ from typing import Any, Dict
|
|
| 11 |
|
| 12 |
import pypdfium2
|
| 13 |
import streamlit as st
|
|
|
|
| 14 |
|
| 15 |
from marker.converters.pdf import PdfConverter
|
| 16 |
from marker.models import create_model_dict
|
|
@@ -109,10 +112,12 @@ with col1:
|
|
| 109 |
st.image(pil_image, caption="PDF file (preview)", use_container_width=True)
|
| 110 |
|
| 111 |
page_range = st.sidebar.text_input("Page range to parse, comma separated like 0,5-10,20", value=f"{page_number}-{page_number}")
|
| 112 |
-
force_ocr = st.sidebar.checkbox("Force OCR on all pages", help="Force OCR on all pages, even if they are images", value=False)
|
| 113 |
output_format = st.sidebar.selectbox("Output format", ["markdown", "json", "html"], index=0)
|
| 114 |
run_marker = st.sidebar.button("Run Marker")
|
| 115 |
|
|
|
|
|
|
|
|
|
|
| 116 |
if not run_marker:
|
| 117 |
st.stop()
|
| 118 |
|
|
@@ -121,7 +126,14 @@ with tempfile.NamedTemporaryFile(suffix=".pdf") as temp_pdf:
|
|
| 121 |
temp_pdf.write(in_file.getvalue())
|
| 122 |
temp_pdf.seek(0)
|
| 123 |
filename = temp_pdf.name
|
| 124 |
-
rendered = convert_pdf(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 125 |
|
| 126 |
text, ext, images = text_from_rendered(rendered)
|
| 127 |
with col2:
|
|
@@ -133,3 +145,14 @@ with col2:
|
|
| 133 |
elif output_format == "html":
|
| 134 |
st.html(text)
|
| 135 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
|
| 3 |
+
from marker.settings import settings
|
| 4 |
+
|
| 5 |
os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"
|
| 6 |
os.environ["IN_STREAMLIT"] = "true"
|
| 7 |
|
|
|
|
| 13 |
|
| 14 |
import pypdfium2
|
| 15 |
import streamlit as st
|
| 16 |
+
from PIL import Image
|
| 17 |
|
| 18 |
from marker.converters.pdf import PdfConverter
|
| 19 |
from marker.models import create_model_dict
|
|
|
|
| 112 |
st.image(pil_image, caption="PDF file (preview)", use_container_width=True)
|
| 113 |
|
| 114 |
page_range = st.sidebar.text_input("Page range to parse, comma separated like 0,5-10,20", value=f"{page_number}-{page_number}")
|
|
|
|
| 115 |
output_format = st.sidebar.selectbox("Output format", ["markdown", "json", "html"], index=0)
|
| 116 |
run_marker = st.sidebar.button("Run Marker")
|
| 117 |
|
| 118 |
+
force_ocr = st.sidebar.checkbox("Force OCR", help="Force OCR on all pages", value=False)
|
| 119 |
+
debug = st.sidebar.checkbox("Debug", help="Show debug information", value=False)
|
| 120 |
+
|
| 121 |
if not run_marker:
|
| 122 |
st.stop()
|
| 123 |
|
|
|
|
| 126 |
temp_pdf.write(in_file.getvalue())
|
| 127 |
temp_pdf.seek(0)
|
| 128 |
filename = temp_pdf.name
|
| 129 |
+
rendered = convert_pdf(
|
| 130 |
+
filename,
|
| 131 |
+
page_range=page_range,
|
| 132 |
+
force_ocr=force_ocr,
|
| 133 |
+
output_format=output_format,
|
| 134 |
+
output_dir=settings.DEBUG_DATA_FOLDER if debug else None,
|
| 135 |
+
debug=debug
|
| 136 |
+
)
|
| 137 |
|
| 138 |
text, ext, images = text_from_rendered(rendered)
|
| 139 |
with col2:
|
|
|
|
| 145 |
elif output_format == "html":
|
| 146 |
st.html(text)
|
| 147 |
|
| 148 |
+
if debug:
|
| 149 |
+
with col1:
|
| 150 |
+
debug_data_path = rendered.metadata.get("debug_data_path")
|
| 151 |
+
if debug_data_path:
|
| 152 |
+
pdf_image_path = os.path.join(debug_data_path, f"pdf_page_0.png")
|
| 153 |
+
img = Image.open(pdf_image_path)
|
| 154 |
+
st.image(img, caption="PDF debug image", use_container_width=True)
|
| 155 |
+
layout_image_path = os.path.join(debug_data_path, f"layout_page_0.png")
|
| 156 |
+
img = Image.open(layout_image_path)
|
| 157 |
+
st.image(img, caption="Layout debug image", use_container_width=True)
|
| 158 |
+
|