Expose hybrid OCR in the Python API + fix RapidOCR runner crash

crates/edgeparse-core/src/pdf/raster_table_ocr.rs

@@ -154,9 +154,14 @@ if result is None:
     print('[]')
     raise SystemExit(0)
 
-boxes = getattr(result, 'boxes', []) or []
-txts = getattr(result, 'txts', []) or []
-scores = getattr(result, 'scores', []) or []
+# RapidOCR returns NumPy arrays for these fields; `arr or []` raises
+# "truth value of an array ... is ambiguous", so normalise explicitly.
+boxes = getattr(result, 'boxes', None)
+boxes = [] if boxes is None else list(boxes)
+txts = getattr(result, 'txts', None)
+txts = [] if txts is None else list(txts)
+scores = getattr(result, 'scores', None)
+scores = [] if scores is None else list(scores)
 out = []
 for box, text, score in zip(boxes, txts, scores):
     if not text or not str(text).strip():

crates/edgeparse-python/src/lib.rs

@@ -2,7 +2,8 @@ use pyo3::exceptions::PyRuntimeError;
 use pyo3::prelude::*;
 
 use edgeparse_core::api::config::{
-    ImageFormat, ImageOutput, OutputFormat, ProcessingConfig, ReadingOrder, TableMethod,
+    HybridBackend, HybridMode, ImageFormat, ImageOutput, OutputFormat, ProcessingConfig,
+    ReadingOrder, TableMethod,
 };
 use edgeparse_core::output;
 
@@ -18,6 +19,11 @@ use std::path::Path;
 ///   reading_order: Reading order algorithm — "xycut" or "off". Default: "xycut".
 ///   table_method: Table detection method — "default" or "cluster". Default: "default".
 ///   image_output: Image output mode — "off", "embedded", or "external". Default: "off".
+///   hybrid: Hybrid OCR backend — "off" or "docling-fast". Default: "off".
+///           "docling-fast" enables OCR recovery of image-based text and tables
+///           (requires pdftoppm, tesseract, and optionally rapidocr on the host).
+///   hybrid_mode: Hybrid triage — "auto" (route complex pages) or "full" (all
+///           pages). Only applies when hybrid is enabled. Default: "auto".
 ///
 /// Returns:
 ///   The extracted content as a string in the requested format.
@@ -31,7 +37,10 @@ use std::path::Path;
     reading_order = "xycut",
     table_method = "default",
     image_output = "off",
+    hybrid = "off",
+    hybrid_mode = "auto",
 ))]
+#[allow(clippy::too_many_arguments)]
 fn convert(
     input_path: &str,
     format: &str,
@@ -40,6 +49,8 @@ fn convert(
     reading_order: &str,
     table_method: &str,
     image_output: &str,
+    hybrid: &str,
+    hybrid_mode: &str,
 ) -> PyResult<String> {
     let pdf_path = Path::new(input_path);
     if !pdf_path.exists() {
@@ -78,6 +89,14 @@ fn convert(
             _ => ImageOutput::Off,
         },
         image_format: ImageFormat::Png,
+        hybrid: match hybrid {
+            "docling-fast" => HybridBackend::DoclingFast,
+            _ => HybridBackend::Off,
+        },
+        hybrid_mode: match hybrid_mode {
+            "full" => HybridMode::Full,
+            _ => HybridMode::Auto,
+        },
         ..ProcessingConfig::default()
     };
 
@@ -118,6 +137,8 @@ fn convert(
 ///   format: Output format — "markdown", "json", "html", or "text". Default: "markdown".
 ///   pages: Optional page range string, e.g. "1,3,5-7".
 ///   password: Optional password for encrypted PDFs.
+///   hybrid: Hybrid OCR backend — "off" or "docling-fast". Default: "off".
+///   hybrid_mode: Hybrid triage — "auto" or "full". Default: "auto".
 ///
 /// Returns:
 ///   The path to the created output file.
@@ -129,16 +150,20 @@ fn convert(
     format = "markdown",
     pages = None,
     password = None,
+    hybrid = "off",
+    hybrid_mode = "auto",
 ))]
 fn convert_file(
     input_path: &str,
     output_dir: &str,
     format: &str,
     pages: Option<&str>,
     password: Option<&str>,
+    hybrid: &str,
+    hybrid_mode: &str,
 ) -> PyResult<String> {
     let content = convert(
-        input_path, format, pages, password, "xycut", "default", "off",
+        input_path, format, pages, password, "xycut", "default", "off", hybrid, hybrid_mode,
     )?;
 
     let out_dir = Path::new(output_dir);

sdks/python/edgeparse/__init__.py

@@ -23,6 +23,8 @@ def convert(
     reading_order: str = "xycut",
     table_method: str = "default",
     image_output: str = "off",
+    hybrid: str = "off",
+    hybrid_mode: str = "auto",
 ) -> str:
     """Convert a PDF file and return the extracted content as a string.
 
@@ -43,6 +45,14 @@ def convert(
         Table detection method. ``"default"`` or ``"cluster"``.
     image_output:
         Image output mode. ``"off"`` (default), ``"embedded"``, or ``"external"``.
+    hybrid:
+        Hybrid OCR backend. ``"off"`` (default) or ``"docling-fast"``.
+        ``"docling-fast"`` enables OCR recovery of image-based text and tables
+        and requires ``pdftoppm`` and ``tesseract`` (and optionally ``rapidocr``)
+        to be available on the host.
+    hybrid_mode:
+        Hybrid triage mode. ``"auto"`` (default) routes complex pages to OCR;
+        ``"full"`` runs OCR on every page. Only applies when ``hybrid`` is enabled.
 
     Returns
     -------
@@ -57,6 +67,8 @@ def convert(
         reading_order=reading_order,
         table_method=table_method,
         image_output=image_output,
+        hybrid=hybrid,
+        hybrid_mode=hybrid_mode,
     )
 
 
@@ -67,6 +79,8 @@ def convert_file(
     format: str = "markdown",
     pages: Optional[str] = None,
     password: Optional[str] = None,
+    hybrid: str = "off",
+    hybrid_mode: str = "auto",
 ) -> str:
     """Convert a PDF file and write the output to a file.
 
@@ -83,6 +97,10 @@ def convert_file(
         Optional page range string, e.g. ``"1,3,5-7"``.
     password:
         Optional password for encrypted PDFs.
+    hybrid:
+        Hybrid OCR backend. ``"off"`` (default) or ``"docling-fast"``.
+    hybrid_mode:
+        Hybrid triage mode. ``"auto"`` (default) or ``"full"``.
 
     Returns
     -------
@@ -95,6 +113,8 @@ def convert_file(
         format=format,
         pages=pages,
         password=password,
+        hybrid=hybrid,
+        hybrid_mode=hybrid_mode,
     )
 
 

sdks/python/edgeparse/_types.py

@@ -16,3 +16,9 @@
 
 # Valid image output modes
 IMAGE_OUTPUTS = ("off", "embedded", "external")
+
+# Valid hybrid OCR backends
+HYBRID_BACKENDS = ("off", "docling-fast")
+
+# Valid hybrid triage modes
+HYBRID_MODES = ("auto", "full")