glm-ocr-fixed / app.py
rehan953's picture
Update app.py
bcc62dd verified
Raw
History Blame
7.8 kB
"""
GLM-OCR Hugging Face Space app with client-side header/footer fallback for MaaS.
Works for every PDF and every image: uses API bboxes when available, else minimal band.
"""
import os
import re
import tempfile
import yaml
import gradio as gr
import glmocr
GLMOCR_BASE = os.path.dirname(glmocr.__file__)
CONFIG_PATH = os.path.join(GLMOCR_BASE, "config.yaml")
FORMATTER_PATH = os.path.join(GLMOCR_BASE, "postprocess", "result_formatter.py")
DEFAULT_ZONE_FRAC = float(os.environ.get("GLMOCR_DEFAULT_ZONE_FRAC", "0.08"))
# ---------------------------------------------------------------------------
# 1. Config: include headers/footers (remove from abandon lists)
# ---------------------------------------------------------------------------
with open(CONFIG_PATH, "r") as f:
config = yaml.safe_load(f)
config["pipeline"]["maas"]["enabled"] = True
config["pipeline"]["maas"]["api_key"] = "4570c28bdea5493c9efae9dae68edc66.sGbA9DLlcX1GlvqV"
to_remove = {"header", "footer"}
for key in ("layout", "result_formatter"):
if key not in config.get("pipeline", {}):
continue
section = config["pipeline"][key]
if key == "layout" and "label_task_mapping" in section:
abandon = section["label_task_mapping"].get("abandon")
if isinstance(abandon, list):
section["label_task_mapping"]["abandon"] = [x for x in abandon if x not in to_remove]
elif key == "result_formatter" and "abandon" in section:
abandon = section["abandon"]
if isinstance(abandon, list):
section["abandon"] = [x for x in abandon if x not in to_remove]
with open(CONFIG_PATH, "w") as f:
yaml.dump(config, f, default_flow_style=False, sort_keys=False)
# ---------------------------------------------------------------------------
# 2. result_formatter: stop stripping header/footer in postprocess
# ---------------------------------------------------------------------------
with open(FORMATTER_PATH, "r") as f:
source = f.read()
for label in ('"header"', "'header'", '"footer"', "'footer'", '"doc_header"', "'doc_header'", '"doc_footer"', "'doc_footer'"):
source = re.sub(r",\s*" + re.escape(label), "", source)
source = re.sub(re.escape(label) + r"\s*,", "", source)
source = re.sub(re.escape(label), "", source)
with open(FORMATTER_PATH, "w") as f:
f.write(source)
def get_header_footer_zones(regions, norm_height=1000):
if not regions:
return None, None
y_tops, y_bottoms = [], []
for r in regions:
bbox = r.get("bbox_2d") if isinstance(r, dict) else getattr(r, "bbox_2d", None)
if bbox and len(bbox) >= 4:
y_tops.append(bbox[1])
y_bottoms.append(bbox[3])
if not y_tops:
return None, None
return min(y_tops) / norm_height, max(y_bottoms) / norm_height
def extract_zone_text_pdf(pdf_path, page_num, y_start_frac, y_end_frac):
try:
import pymupdf as fitz
doc = fitz.open(pdf_path)
page = doc[page_num]
h, w = page.rect.height, page.rect.width
rect = fitz.Rect(0, h * y_start_frac, w, h * y_end_frac)
text = page.get_text(clip=rect).strip()
doc.close()
return text
except Exception:
return ""
def ocr_zone(image_path, y_start_frac, y_end_frac):
try:
from PIL import Image
from glmocr import parse
img = Image.open(image_path).convert("RGB")
w, h = img.size
y0 = max(0, int(h * y_start_frac))
y1 = min(h, int(h * y_end_frac))
if y1 <= y0:
return ""
crop = img.crop((0, y0, w, y1))
fd, path = tempfile.mkstemp(suffix=".png")
os.close(fd)
crop.save(path)
try:
out = parse(path)
if not isinstance(out, list):
out = [out]
if out and getattr(out[0], "markdown_result", None):
return (out[0].markdown_result or "").strip()
finally:
try:
os.unlink(path)
except Exception:
pass
except Exception:
pass
return ""
def get_page_md_and_regions(page_result):
md = ""
if hasattr(page_result, "markdown_result") and page_result.markdown_result:
md = (page_result.markdown_result or "").strip()
regions = []
if hasattr(page_result, "json_result"):
jr = page_result.json_result
if isinstance(jr, list) and len(jr) > 0:
r = jr[0] if isinstance(jr[0], list) else jr
if isinstance(r, list):
regions = r
elif isinstance(r, dict) and "regions" in r:
regions = r.get("regions") or []
return md, regions
def run_ocr(uploaded_file):
if uploaded_file is None:
return "Please upload a file."
try:
from glmocr import parse
import pymupdf as fitz
path = uploaded_file.name if hasattr(uploaded_file, "name") else str(uploaded_file)
is_pdf = path.lower().endswith(".pdf")
if is_pdf:
doc = fitz.open(path)
page_images = []
for i in range(len(doc)):
pix = doc[i].get_pixmap(matrix=fitz.Matrix(1.5, 1.5), alpha=False)
img_path = os.path.join(tempfile.gettempdir(), f"maas_page_{i}.png")
pix.save(img_path)
page_images.append(img_path)
doc.close()
results = parse(page_images)
else:
page_images = [path]
results = parse(path)
if not isinstance(results, list):
results = [results]
all_pages = []
for page_num, page_result in enumerate(results):
page_md, regions = get_page_md_and_regions(page_result)
header_end_frac, footer_start_frac = get_header_footer_zones(regions, 1000)
he = header_end_frac if header_end_frac is not None else DEFAULT_ZONE_FRAC
fs = footer_start_frac if footer_start_frac is not None else (1.0 - DEFAULT_ZONE_FRAC)
if he <= 0 and header_end_frac is not None:
he = 0
if fs >= 1.0 and footer_start_frac is not None:
fs = 1.0
parts = []
if he > 0 and page_num < len(page_images):
img_path = page_images[page_num]
hdr = ""
if is_pdf:
hdr = extract_zone_text_pdf(path, page_num, 0, he)
if not (hdr and hdr.strip()):
hdr = ocr_zone(img_path, 0, he)
if hdr and hdr.strip():
parts.append(hdr.strip())
if page_md:
parts.append(page_md)
if fs < 1.0 and page_num < len(page_images):
img_path = page_images[page_num]
ftr = ""
if is_pdf:
ftr = extract_zone_text_pdf(path, page_num, fs, 1.0)
if not (ftr and ftr.strip()):
ftr = ocr_zone(img_path, fs, 1.0)
if ftr and ftr.strip():
parts.append(ftr.strip())
if parts:
all_pages.append("\n\n".join(parts))
return "\n\n---\n\n".join(all_pages) if all_pages else "(No content)"
except Exception as e:
import traceback
return f"Error: {e}\n\n{traceback.format_exc()}"
with gr.Blocks(title="GLM-OCR") as demo:
gr.Markdown("# GLM-OCR\nUpload a PDF or image. Headers and footers included.")
file_in = gr.File(label="Upload PDF or image", file_types=[".pdf", ".png", ".jpg", ".jpeg", ".tiff", ".bmp"])
run_btn = gr.Button("Run OCR", variant="primary")
out = gr.Textbox(lines=40, label="Output")
run_btn.click(fn=run_ocr, inputs=file_in, outputs=out)
if __name__ == "__main__":
demo.launch()