Upload app.py with huggingface_hub

app.py

@@ -3,7 +3,6 @@ import yaml
 import re
 import os
 
-# Paths from installed glmocr package
 import glmocr
 GLMOCR_BASE    = os.path.dirname(glmocr.__file__)
 config_path    = os.path.join(GLMOCR_BASE, "config.yaml")
@@ -13,8 +12,8 @@ formatter_path = os.path.join(GLMOCR_BASE, "postprocess", "result_formatter.py")
 with open(config_path, "r") as f:
     config = yaml.safe_load(f)
 
-config["pipeline"]["maas"]["enabled"] = True
-config["pipeline"]["maas"]["api_key"] = "4570c28bdea5493c9efae9dae68edc66.sGbA9DLlcX1GlvqV"
+config["pipeline"]["maas"]["enabled"]  = True
+config["pipeline"]["maas"]["api_key"]  = "4570c28bdea5493c9efae9dae68edc66.sGbA9DLlcX1GlvqV"
 
 config["pipeline"]["result_formatter"]["abandon"] = [
     "number", "footnote", "aside_text",
@@ -50,9 +49,8 @@ print("✅ result_formatter.py fixed")
 ABANDON = set(config["pipeline"]["result_formatter"]["abandon"])
 
 # ── STEP 3: Extract header/footer from PDF text layer ─────────
-# The MaaS API only returns 'text' and 'table' labels — it never
-# labels anything as 'header' or 'footer'. So we extract those
-# directly from the PDF text layer using PyMuPDF instead.
+# The MaaS API only returns 'text' and 'table' labels.
+# Headers and footers are extracted directly from the PDF using PyMuPDF.
 def extract_pdf_headers(path):
     import fitz
     page_headers = []
@@ -60,9 +58,7 @@ def extract_pdf_headers(path):
     for page in doc:
         page_rect   = page.rect
         page_height = page_rect.height
-        # Top 12% = header area
         header_rect = fitz.Rect(0, 0, page_rect.width, page_height * 0.12)
-        # Bottom 12% = footer area
         footer_rect = fitz.Rect(0, page_height * 0.88, page_rect.width, page_height)
         header_text = page.get_text(clip=header_rect).strip()
         footer_text = page.get_text(clip=footer_rect).strip()
@@ -70,7 +66,29 @@ def extract_pdf_headers(path):
     doc.close()
     return page_headers
 
-# ── STEP 4: Main OCR function ─────────────────────────────────
+# ── STEP 4: Parse one page result into regions list ───────────
+# result from parse() is a LIST of PipelineResult objects (one per page)
+# Each PipelineResult has:
+#   .json_result  → list of region dicts with 'label' and 'content'
+#   .markdown_result → plain markdown string
+def get_page_regions(page_result):
+    # Try json_result first (structured regions)
+    if hasattr(page_result, "json_result"):
+        jr = page_result.json_result
+        if isinstance(jr, list) and len(jr) > 0:
+            # json_result is a list of lists (one list per page image)
+            # since we pass one image per call, take index 0
+            regions = jr[0] if isinstance(jr[0], list) else jr
+            if isinstance(regions, list):
+                return regions, "json"
+    # Fallback to markdown_result
+    if hasattr(page_result, "markdown_result"):
+        md = page_result.markdown_result
+        if md and isinstance(md, str) and md.strip():
+            return md.strip(), "raw"
+    return [], "empty"
+
+# ── STEP 5: Main OCR function ─────────────────────────────────
 def run_ocr(uploaded_file):
     if uploaded_file is None:
         return "Please upload a file.", "No regions detected."
@@ -80,7 +98,6 @@ def run_ocr(uploaded_file):
 
         path = uploaded_file.name if hasattr(uploaded_file, "name") else str(uploaded_file)
 
-        # Convert PDF to images + extract headers from text layer
         if path.lower().endswith(".pdf"):
             import fitz
             pdf_headers = extract_pdf_headers(path)
@@ -92,30 +109,26 @@ def run_ocr(uploaded_file):
                 pix.save(img_path)
                 page_images.append(img_path)
             doc.close()
-            result = parse(page_images)
+            # parse() returns a LIST — one PipelineResult per page
+            results = parse(page_images)
+            if not isinstance(results, list):
+                results = [results]
         else:
             pdf_headers = []
-            result = parse(path)
-
-        # Parse result
-        pages_data = result.json_result if hasattr(result, "json_result") else []
-        if not isinstance(pages_data, list):
-            pages_data = [pages_data] if pages_data else []
+            results = parse(path)
+            if not isinstance(results, list):
+                results = [results]
 
         total_headers = 0
         total_footers = 0
         all_markdown  = []
         all_summary   = []
 
-        for page_num, page_regions in enumerate(pages_data):
-            if not isinstance(page_regions, list):
-                page_regions = [page_regions] if page_regions else []
-
+        for page_num, page_result in enumerate(results):
             page_md      = []
-            page_summary = [f"── PAGE {page_num + 1} of {len(pages_data)} ──"]
+            page_summary = [f"── PAGE {page_num + 1} of {len(results)} ──"]
 
             # Inject header/footer from PDF text layer
-            # (API never returns these as separate labeled regions)
             if pdf_headers and page_num < len(pdf_headers):
                 hdr = pdf_headers[page_num]["header"]
                 ftr = pdf_headers[page_num]["footer"]
@@ -128,38 +141,47 @@ def run_ocr(uploaded_file):
                     page_md.append("<!-- FOOTER -->\n" + ftr)
                     page_summary.append("🟢 FOOTER: " + ftr[:100])
 
-            # Process API regions (text, table, etc.)
-            for region in page_regions:
-                if not isinstance(region, dict):
-                    continue
-                label   = region.get("label", "text")
-                content = str(region.get("content", ""))
-
-                if label in ABANDON:
-                    continue
-
-                # API returns 'header'/'footer' labels in some docs
-                # Keep them too in case the API does label them
-                if label == "header":
-                    total_headers += 1
-                    page_summary.append("🔵 HEADER (API): " + content[:100])
-                    page_md.append("<!-- HEADER -->\n" + content)
-                elif label == "footer":
-                    total_footers += 1
-                    page_summary.append("🟢 FOOTER (API): " + content[:100])
-                    page_md.append("<!-- FOOTER -->\n" + content)
-                else:
-                    page_summary.append("[" + label + "]: " + content[:100])
-                    page_md.append(content)
+            # Get regions from this page result
+            data, dtype = get_page_regions(page_result)
+
+            if dtype == "json":
+                for region in data:
+                    if not isinstance(region, dict):
+                        continue
+                    label   = region.get("label", "text")
+                    content = str(region.get("content", ""))
+
+                    if label in ABANDON:
+                        continue
+
+                    if label == "header":
+                        total_headers += 1
+                        page_summary.append("🔵 HEADER (API): " + content[:100])
+                        page_md.append("<!-- HEADER -->\n" + content)
+                    elif label == "footer":
+                        total_footers += 1
+                        page_summary.append("🟢 FOOTER (API): " + content[:100])
+                        page_md.append("<!-- FOOTER -->\n" + content)
+                    else:
+                        page_summary.append("[" + label + "]: " + content[:100])
+                        page_md.append(content)
+
+            elif dtype == "raw":
+                page_md.append(data)
+                page_summary.append("[markdown output]")
+                page_summary.append("Content length: " + str(len(data)) + " chars")
+
+            else:
+                page_summary.append("[empty page]")
 
             all_markdown.append("\n\n".join(page_md))
             all_summary.extend(page_summary)
             all_summary.append("")
 
         summary = (
-            "Total pages   : " + str(len(pages_data)) + "\n"
-            "Headers found : " + str(total_headers) + "\n"
-            "Footers found : " + str(total_footers) + "\n"
+            "Total pages   : " + str(len(results)) + "\n"
+            + "Headers found : " + str(total_headers) + "\n"
+            + "Footers found : " + str(total_footers) + "\n"
             + "─" * 40 + "\n"
             + "\n".join(all_summary)
         )
@@ -171,7 +193,7 @@ def run_ocr(uploaded_file):
         import traceback
         return "Error: " + str(e) + "\n\n" + traceback.format_exc(), "Failed."
 
-# ── STEP 5: Gradio UI ─────────────────────────────────────────
+# ── STEP 6: Gradio UI ─────────────────────────────────────────
 with gr.Blocks(title="GLM-OCR — MaaS (Header & Footer)") as demo:
     gr.Markdown("""
     # 🔍 GLM-OCR — Zhipu MaaS
@@ -196,4 +218,4 @@ with gr.Blocks(title="GLM-OCR — MaaS (Header & Footer)") as demo:
     run_btn.click(fn=run_ocr, inputs=file_input,
                   outputs=[markdown_out, regions_out])
 
-demo.launch()
+demo.launch()