kreuzberg 4.6.1 → 4.6.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 7f2441d44e9083d36f9f0e60b2c155624ad251f97c6f2ca629bd3ef33aeeeb46
-  data.tar.gz: b443884a5e4bfffa1bc916928ae97997e1ca21cdfb80e76b845c454c0aa67256
+  metadata.gz: f087c152499deb621223ba54a8ca450bb0510da3b43c880f3969e33ec4b4d5e5
+  data.tar.gz: dba38092babe378ec93e0dd4f307ab44c10519623302fdf14023a52eff6549b2
 SHA512:
-  metadata.gz: e238336b3ceae6d2bed4bd530d393f32473a7a88f68f7e28a6300ffe2f7cadc5aee0ac834fa5712223810f9cf1f7a68a363b7e96079060565e1c9e10af9f8114
-  data.tar.gz: 48eae2f77c78ef1b9794ac3dde62c785b32ab99253b6ab7ce8da9d8bba8a57f209fb5fc6353a0bc8f5771bd825fc3b3b29652149133bd0ba1dc29e58101185bb
+  metadata.gz: 2e2cb3a3636555ac5f7d54bf473e36872efef5cf1796c667f727fa8622792a5cec009799d35c5ccb776997b7c39a0e5d0745266c7c5e5604d22b4e3e3e2f5e1f
+  data.tar.gz: b032174ab9b1366d3d05082ef6cfe3b1b700ff66d265fb28353d69b581aee6989688dd8e01b8c516f7046a8968ca86edd8a80cb72cb48477607eb8104d7facb3

data/README.md

@@ -22,7 +22,7 @@
     <img src="https://img.shields.io/maven-central/v/dev.kreuzberg/kreuzberg?label=Java&color=007ec6" alt="Java">
   </a>
   <a href="https://github.com/kreuzberg-dev/kreuzberg/releases">
-    <img src="https://img.shields.io/github/v/tag/kreuzberg-dev/kreuzberg?label=Go&color=007ec6&filter=v4.6.1" alt="Go">
+    <img src="https://img.shields.io/github/v/tag/kreuzberg-dev/kreuzberg?label=Go&color=007ec6&filter=v4.6.2" alt="Go">
   </a>
   <a href="https://www.nuget.org/packages/Kreuzberg/">
     <img src="https://img.shields.io/nuget/v/Kreuzberg?label=C%23&color=007ec6" alt="C#">

data/ext/kreuzberg_rb/native/Cargo.lock

@@ -499,6 +499,15 @@ dependencies = [
  "generic-array",
 ]
 
+[[package]]
+name = "block-buffer"
+version = "0.12.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cdd35008169921d80bc60d3d0ab416eecb028c4cd653352907921d95084790be"
+dependencies = [
+ "hybrid-array",
+]
+
 [[package]]
 name = "block-padding"
 version = "0.3.3"
@@ -726,7 +735,7 @@ version = "0.4.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "773f3b9af64447d2ce9850330c473515014aa235e6a783b02db81ff39e4a3dad"
 dependencies = [
- "crypto-common",
+ "crypto-common 0.1.7",
  "inout",
 ]
 
@@ -877,6 +886,12 @@ dependencies = [
  "web-sys",
 ]
 
+[[package]]
+name = "const-oid"
+version = "0.10.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a6ef517f0926dd24a1582492c791b6a4818a4d94e789a334894aa15b0d12f55c"
+
 [[package]]
 name = "const-random"
 version = "0.1.18"
@@ -1034,6 +1049,15 @@ dependencies = [
  "typenum",
 ]
 
+[[package]]
+name = "crypto-common"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "77727bb15fa921304124b128af125e7e3b968275d1b108b379190264f4423710"
+dependencies = [
+ "hybrid-array",
+]
+
 [[package]]
 name = "ctor"
 version = "0.6.3"
@@ -1232,8 +1256,19 @@ version = "0.10.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9ed9a281f7bc9b7576e61468ba615a66a5c8cfdff42420a70aa82701a3b1e292"
 dependencies = [
- "block-buffer",
- "crypto-common",
+ "block-buffer 0.10.4",
+ "crypto-common 0.1.7",
+]
+
+[[package]]
+name = "digest"
+version = "0.11.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4850db49bf08e663084f7fb5c87d202ef91a3907271aff24a94eb97ff039153c"
+dependencies = [
+ "block-buffer 0.12.0",
+ "const-oid",
+ "crypto-common 0.2.1",
 ]
 
 [[package]]
@@ -2061,6 +2096,15 @@ version = "1.0.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "df3b46402a9d5adb4c86a0cf463f42e19994e3ee891101b1841f30a545cb49a9"
 
+[[package]]
+name = "hybrid-array"
+version = "0.4.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8655f91cd07f2b9d0c24137bd650fe69617773435ee5ec83022377777ce65ef1"
+dependencies = [
+ "typenum",
+]
+
 [[package]]
 name = "hyper"
 version = "1.8.1"
@@ -2612,7 +2656,7 @@ dependencies = [
 
 [[package]]
 name = "kreuzberg"
-version = "4.6.0"
+version = "4.6.2"
 dependencies = [
  "ahash",
  "async-trait",
@@ -2626,6 +2670,7 @@ dependencies = [
  "calamine",
  "cfb 0.14.0",
  "chardetng",
+ "chrono",
  "dashmap",
  "dbase",
  "encoding_rs",
@@ -2658,6 +2703,7 @@ dependencies = [
  "opentelemetry_sdk",
  "org",
  "ort",
+ "outlook-pst",
  "parking_lot",
  "pastey 0.2.1",
  "pkg-config",
@@ -2673,9 +2719,10 @@ dependencies = [
  "serde_json",
  "serde_yaml_ng",
  "sevenz-rust2",
- "sha2",
+ "sha2 0.11.0",
  "snap",
  "tar",
+ "tempfile",
  "text-splitter",
  "thiserror 2.0.18",
  "tiff",
@@ -2696,7 +2743,7 @@ dependencies = [
 
 [[package]]
 name = "kreuzberg-ffi"
-version = "4.6.0"
+version = "4.6.2"
 dependencies = [
  "ahash",
  "async-trait",
@@ -2712,7 +2759,7 @@ dependencies = [
 
 [[package]]
 name = "kreuzberg-paddle-ocr"
-version = "4.6.0"
+version = "4.6.2"
 dependencies = [
  "geo-clipper",
  "geo-types",
@@ -2726,7 +2773,7 @@ dependencies = [
 
 [[package]]
 name = "kreuzberg-pdfium-render"
-version = "4.6.0"
+version = "4.6.2"
 dependencies = [
  "bitflags",
  "bytemuck",
@@ -2749,7 +2796,7 @@ dependencies = [
 
 [[package]]
 name = "kreuzberg-rb"
-version = "4.6.0"
+version = "4.6.2"
 dependencies = [
  "async-trait",
  "html-to-markdown-rs",
@@ -2766,7 +2813,7 @@ dependencies = [
 
 [[package]]
 name = "kreuzberg-tesseract"
-version = "4.6.0"
+version = "4.6.2"
 dependencies = [
  "cc",
  "cmake",
@@ -2931,7 +2978,7 @@ dependencies = [
  "rand 0.10.0",
  "rangemap",
  "rayon",
- "sha2",
+ "sha2 0.10.9",
  "stringprep",
  "thiserror 2.0.18",
  "time",
@@ -3068,7 +3115,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d89e7ee0cfbedfc4da3340218492196241d89eefb6dab27de5df917a6d2e78cf"
 dependencies = [
  "cfg-if",
- "digest",
+ "digest 0.10.7",
 ]
 
 [[package]]
@@ -3540,6 +3587,17 @@ dependencies = [
  "ureq 3.3.0",
 ]
 
+[[package]]
+name = "outlook-pst"
+version = "1.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "299eace9d895cc31490927d7de779f8e8f3deb5fb6bd1e68eb192aa1da19bd47"
+dependencies = [
+ "byteorder",
+ "thiserror 2.0.18",
+ "tracing",
+]
+
 [[package]]
 name = "parking_lot"
 version = "0.12.5"
@@ -4652,7 +4710,7 @@ dependencies = [
  "js-sys",
  "lzma-rust2 0.16.2",
  "ppmd-rust",
- "sha2",
+ "sha2 0.10.9",
  "wasm-bindgen",
 ]
 
@@ -4664,7 +4722,18 @@ checksum = "a7507d819769d01a365ab707794a4084392c824f54a7a6a7862f8c3d0892b283"
 dependencies = [
  "cfg-if",
  "cpufeatures 0.2.17",
- "digest",
+ "digest 0.10.7",
+]
+
+[[package]]
+name = "sha2"
+version = "0.11.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "446ba717509524cb3f22f17ecc096f10f4822d76ab5c0b9822c5f9c284e825f4"
+dependencies = [
+ "cfg-if",
+ "cpufeatures 0.3.0",
+ "digest 0.11.2",
 ]
 
 [[package]]

data/ext/kreuzberg_rb/native/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "kreuzberg-rb"
-version = "4.6.1"
+version = "4.6.2"
 edition = "2024"
 rust-version = "1.91"
 authors = ["Na'aman Hirschfeld <naaman@kreuzberg.dev>"]
@@ -36,6 +36,7 @@ kreuzberg = { path = "../../../vendor/kreuzberg", default-features = false, feat
     "chunking",
     "chunking-tokenizers",
     "embeddings",
+    "ort-bundled",
     "quality",
     "keywords",
     "api",

data/ext/kreuzberg_rb/native/src/extraction.rs

@@ -58,6 +58,36 @@ pub fn extract_file(args: &[Value]) -> Result<RHash, Error> {
     extraction_result_to_ruby(&ruby, result)
 }
 
+/// Iterate over PDF pages, yielding (page_index, png_bytes) per page to a Ruby block.
+pub fn render_pdf_pages_iter(path: String, dpi: i32) -> Result<(), Error> {
+    let ruby = Ruby::get().expect("Ruby not initialized");
+    let dpi_opt = if dpi <= 0 { None } else { Some(dpi) };
+
+    let iter = kreuzberg::pdf::PdfPageIterator::from_file(&path, dpi_opt, None)
+        .map_err(|e| kreuzberg_error(e.into()))?;
+
+    for result in iter {
+        let (page_index, png_bytes) = result.map_err(|e| kreuzberg_error(e.into()))?;
+        let rb_index = ruby.integer_from_i64(page_index as i64);
+        let rb_bytes = ruby.str_from_slice(&png_bytes);
+        let _: magnus::Value = ruby.yield_values((rb_index, rb_bytes))?;
+    }
+
+    Ok(())
+}
+
+/// Render a single PDF page to PNG bytes.
+pub fn native_render_pdf_page(path: String, page_index: i64, dpi: i64) -> Result<Vec<u8>, Error> {
+    if page_index < 0 {
+        return Err(crate::error_handling::runtime_error("page_index must be non-negative"));
+    }
+    let pdf_bytes = std::fs::read(&path)
+        .map_err(|e| crate::error_handling::runtime_error(format!("Failed to read file: {}", e)))?;
+    let dpi_opt = if dpi <= 0 { None } else { Some(dpi as i32) };
+    kreuzberg::pdf::render_pdf_page_to_png(&pdf_bytes, page_index as usize, dpi_opt, None)
+        .map_err(|e| kreuzberg_error(e.into()))
+}
+
 /// Extract content from bytes (asynchronous)
 pub fn extract_bytes(args: &[Value]) -> Result<RHash, Error> {
     let ruby = Ruby::get().expect("Ruby not initialized");

data/ext/kreuzberg_rb/native/src/lib.rs

@@ -23,7 +23,7 @@ pub use gc_guarded_value::GcGuardedValue;
 pub use helpers::{get_kw, set_hash_entry, json_value_to_ruby, ruby_value_to_json, cache_root_dir, cache_directories};
 pub use config::parse_extraction_config;
 pub use result::extraction_result_to_ruby;
-pub use extraction::{extract_file_sync, extract_bytes_sync, extract_file, extract_bytes};
+pub use extraction::{extract_file_sync, extract_bytes_sync, extract_file, extract_bytes, render_pdf_pages_iter, native_render_pdf_page};
 pub use batch::{
     batch_extract_files_sync, batch_extract_bytes_sync, batch_extract_files, batch_extract_bytes,
 };
@@ -442,6 +442,10 @@ fn init(ruby: &Ruby) -> Result<(), Error> {
     module.define_module_function("batch_extract_files", function!(batch_extract_files, -1))?;
     module.define_module_function("batch_extract_bytes", function!(batch_extract_bytes, -1))?;
 
+    // PDF page iterator
+    module.define_module_function("native_render_pdf_pages_iter", function!(render_pdf_pages_iter, 2))?;
+    module.define_module_function("native_render_pdf_page", function!(native_render_pdf_page, 3))?;
+
     // Cache functions
     module.define_module_function("clear_cache", function!(ruby_clear_cache, 0))?;
     module.define_module_function("cache_stats", function!(ruby_cache_stats, 0))?;

data/lib/kreuzberg/extraction_api.rb

@@ -319,6 +319,43 @@ module Kreuzberg
       results
     end
 
+    # Render a single PDF page as a PNG image.
+    #
+    # @param path [String, Pathname] Path to the PDF file
+    # @param page_index [Integer] Zero-based page index
+    # @param dpi [Integer] Rendering resolution (default 150)
+    # @return [String] PNG-encoded binary string
+    # @raise [Errors::IOError] If the file cannot be read
+    # @raise [Errors::ParsingError] If rendering fails
+    def render_pdf_page(path, page_index, dpi: 150)
+      path_str = path.to_s
+      raise ArgumentError, 'page_index must be non-negative' if page_index.negative?
+      raise Errors::IOError, "File not found: #{path_str}" unless File.exist?(path_str)
+
+      native_render_pdf_page(path_str, page_index, dpi)
+    end
+
+    # Iterate over pages of a PDF lazily, yielding each page as it is rendered.
+    #
+    # Each page is rendered via the native FFI iterator, so only one page is in
+    # memory at a time.
+    #
+    # @param path [String, Pathname] Path to the PDF file
+    # @param dpi [Integer] Rendering resolution (default 150)
+    # @yieldparam page_index [Integer] Zero-based page index
+    # @yieldparam png_bytes [String] PNG-encoded binary string for the page
+    # @return [Enumerator] if no block is given
+    # @raise [Errors::IOError] If the file cannot be read
+    # @raise [Errors::ParsingError] If rendering fails
+    def render_pdf_pages_iter(path, dpi: 150, &block)
+      path_str = path.to_s
+      raise Errors::IOError, "File not found: #{path_str}" unless File.exist?(path_str)
+
+      return enum_for(:render_pdf_pages_iter, path, dpi: dpi) unless block
+
+      native_render_pdf_pages_iter(path_str, dpi, &block)
+    end
+
     def normalize_config(config)
       return {} if config.nil?
       return config if config.is_a?(Hash)

data/lib/kreuzberg/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module Kreuzberg
-  VERSION = '4.6.1'
+  VERSION = '4.6.2'
 end

data/spec/binding/render_spec.rb

@@ -0,0 +1,91 @@
+# Hand-written binding-specific edge case tests for PDF rendering.
+# Happy-path render tests are auto-generated from fixtures in e2e/.
+# These tests cover error handling, validation, and lifecycle patterns
+# that vary per language and can't be generated uniformly.
+
+# frozen_string_literal: true
+
+require 'spec_helper'
+
+RSpec.describe 'PDF Rendering' do
+  it 'exposes rendering methods' do
+    expect(Kreuzberg).to respond_to(:render_pdf_page)
+    expect(Kreuzberg).to respond_to(:render_pdf_pages_iter)
+  end
+
+  describe '.render_pdf_page' do
+    it 'raises an error for a nonexistent file' do
+      expect do
+        Kreuzberg.render_pdf_page('/nonexistent/path/to/document.pdf', 0)
+      end.to raise_error(Kreuzberg::Errors::IOError)
+    end
+
+    it 'raises an error for an out-of-bounds page index' do
+      pdf_path = test_document_path('pdf/tiny.pdf')
+      skip 'Test PDF not available' unless File.exist?(pdf_path)
+
+      expect do
+        Kreuzberg.render_pdf_page(pdf_path, 9999)
+      end.to raise_error(StandardError)
+    end
+  end
+
+  describe '.render_pdf_page with negative index' do
+    it 'raises ArgumentError for a negative page index' do
+      pdf_path = test_document_path('pdf/tiny.pdf')
+      skip 'Test PDF not available' unless File.exist?(pdf_path)
+
+      expect do
+        Kreuzberg.render_pdf_page(pdf_path, -1)
+      end.to raise_error(ArgumentError)
+    end
+  end
+
+  describe '.render_pdf_pages_iter' do
+    it 'raises an error for a nonexistent file' do
+      expect do
+        Kreuzberg.render_pdf_pages_iter('/nonexistent/path/to/document.pdf') { |_, _| nil }
+      end.to raise_error(Kreuzberg::Errors::IOError)
+    end
+  end
+
+  describe '.render_pdf_page with empty path' do
+    it 'raises an error for an empty path' do
+      expect do
+        Kreuzberg.render_pdf_page('', 0)
+      end.to raise_error(StandardError)
+    end
+  end
+
+  describe '.render_pdf_pages_iter cleanup' do
+    it 'handles iterator cleanup without fully consuming' do
+      pdf_path = test_document_path('pdf/tiny.pdf')
+      skip 'Test PDF not available' unless File.exist?(pdf_path)
+
+      # Iterate but stop immediately — no crash
+      Kreuzberg.render_pdf_pages_iter(pdf_path) do |_page_index, _png_data|
+        break
+      end
+    end
+  end
+
+  describe '.render_pdf_pages_iter early termination' do
+    it 'returns valid PNG for the first page then stops' do
+      pdf_path = test_document_path('pdf/tiny.pdf')
+      skip 'Test PDF not available' unless File.exist?(pdf_path)
+
+      first_png = nil
+      Kreuzberg.render_pdf_pages_iter(pdf_path) do |page_index, png_data|
+        expect(page_index).to eq(0)
+        expect(png_data).to be_a(String)
+        expect(png_data.bytesize).to be > 8
+        # PNG magic bytes
+        expect(png_data.bytes[0..3]).to eq([0x89, 0x50, 0x4E, 0x47])
+        first_png = png_data
+        break
+      end
+
+      expect(first_png).not_to be_nil
+    end
+  end
+end

data/vendor/Cargo.toml

@@ -2,7 +2,7 @@
 members = ["kreuzberg", "kreuzberg-ffi", "kreuzberg-tesseract", "kreuzberg-paddle-ocr", "kreuzberg-pdfium-render"]
 
 [workspace.package]
-version = "4.6.1"
+version = "4.6.2"
 edition = "2024"
 rust-version = "1.91"
 authors = ["Na'aman Hirschfeld <naaman@kreuzberg.dev>"]
@@ -30,8 +30,8 @@ html-to-markdown-rs = { version = "2.29.0", default-features = false }
 image = { version = "0.25.10", default-features = false }
 itertools = "0.14"
 js-sys = "0.3"
-kreuzberg = { path = "./crates/kreuzberg", version = "4.6.1", default-features = false }
-kreuzberg-ffi = { path = "./crates/kreuzberg-ffi", version = "4.6.1" }
+kreuzberg = { path = "./crates/kreuzberg", version = "4.6.2", default-features = false }
+kreuzberg-ffi = { path = "./crates/kreuzberg-ffi", version = "4.6.2" }
 lazy_static = "1.5.0"
 libc = "0.2.183"
 log = "0.4"

data/vendor/kreuzberg/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "kreuzberg"
-version = "4.6.1"
+version = "4.6.2"
 edition = "2024"
 rust-version = "1.91"
 authors = ["Na'aman Hirschfeld <naaman@kreuzberg.dev>"]
@@ -313,7 +313,7 @@ serde = { version = "1.0.228", features = ["derive"] }
 serde_json = { version = "1.0.149" }
 serde_yaml_ng = "0.10.0"
 sevenz-rust2 = { version = "0.20.2", optional = true }
-sha2 = { version = "0.10", optional = true }
+sha2 = { version = "0.11", optional = true }
 simdutf8 = { version = "0.1", optional = true }
 snap = { version = "1.1", optional = true }
 tar = { version = "0.4.45", optional = true }

data/vendor/kreuzberg/README.md

@@ -18,7 +18,7 @@ High-performance document intelligence library for Rust. Extract text, metadata,
 
 This is the core Rust library that powers the Python, TypeScript, and Ruby bindings.
 
-> **🚀 Version 4.6.1 Release**
+> **🚀 Version 4.6.2 Release**
 > This is a pre-release version. We invite you to test the library and [report any issues](https://github.com/kreuzberg-dev/kreuzberg/issues) you encounter.
 >
 > **Note**: The Rust crate is not currently published to crates.io for this RC. Use git dependencies or language bindings (Python, TypeScript, Ruby) instead.

data/vendor/kreuzberg/src/extraction/pst.rs

@@ -80,6 +80,7 @@ pub fn extract_pst_messages(pst_data: &[u8]) -> Result<(Vec<EmailExtractionResul
 /// occurs when the full PST is first read into a `Vec<u8>` and then written
 /// back out to a tempfile before parsing.
 #[cfg(feature = "email")]
+#[allow(dead_code)]
 pub(crate) fn extract_pst_from_path(
     path: &std::path::Path,
 ) -> Result<(Vec<EmailExtractionResult>, Vec<ProcessingWarning>)> {

data/vendor/kreuzberg/src/extractors/pdf/mod.rs

@@ -294,7 +294,7 @@ impl PdfExtractor {
         let content = &*derotated;
 
         #[cfg(feature = "pdf")]
-        #[allow(unused_variables)]
+        #[allow(unused_variables, unused_mut)]
         let (
             mut pdf_metadata,
             native_text,

data/vendor/kreuzberg/src/extractors/pdf/ocr.rs

@@ -716,7 +716,7 @@ pub(crate) async fn extract_with_ocr(
         // Sequential post-processing for this batch utilizing TATR.
         for offset in 0..batch_count {
             let page_idx = batch_start + offset;
-            let ocr_result = batch_ocr_results[offset].take().expect("OCR result missing for page");
+            let mut ocr_result = batch_ocr_results[offset].take().expect("OCR result missing for page");
             #[cfg(feature = "layout-detection")]
             let height = encoded_batch[offset].3;
 
@@ -731,7 +731,10 @@ pub(crate) async fn extract_with_ocr(
             }
 
             // Accumulate OCR elements from this page.
-            if let Some(ref elems) = ocr_result.ocr_elements {
+            if let Some(ref mut elems) = ocr_result.ocr_elements {
+                for elem in elems.iter_mut() {
+                    elem.page_number = page_idx + 1;
+                }
                 all_ocr_elements.extend(elems.iter().cloned());
             }
 
@@ -741,8 +744,31 @@ pub(crate) async fn extract_with_ocr(
                 && !elements.is_empty()
             {
                 let detection = detections.get(page_idx);
-                let recognized_tables = match (detection, tatr_model.as_mut()) {
-                    (Some(det), Some(model)) => {
+
+                // Scale layout detection bounding boxes from layout-model resolution
+                // (e.g. 640×640) to OCR render resolution so that coordinates are
+                // consistent when passed to recognize_page_tables and detection_to_layout_hints.
+                // `width` and `height` come from the encoded PNG dimensions, i.e. they are
+                // the actual pixel dimensions of the OCR-rendered page image.
+                let ocr_render_width = encoded_batch[offset].2;
+                let ocr_render_height = encoded_batch[offset].3;
+                let scaled_detection: Option<crate::layout::DetectionResult> = detection.map(|det| {
+                    let sx = ocr_render_width as f32 / det.page_width as f32;
+                    let sy = ocr_render_height as f32 / det.page_height as f32;
+                    let mut scaled = det.clone();
+                    scaled.page_width = ocr_render_width;
+                    scaled.page_height = ocr_render_height;
+                    for region in &mut scaled.detections {
+                        region.bbox.x1 *= sx;
+                        region.bbox.y1 *= sy;
+                        region.bbox.x2 *= sx;
+                        region.bbox.y2 *= sy;
+                    }
+                    scaled
+                });
+
+                let recognized_tables = match (scaled_detection.as_ref(), tatr_model.as_mut()) {
+                    (Some(scaled_det), Some(model)) => {
                         // Decode the page image from its PNG for TATR table recognition.
                         // When pre-rendered images are available, use them directly.
                         // Otherwise, decode from the PNG we already encoded.
@@ -757,7 +783,7 @@ pub(crate) async fn extract_with_ocr(
                                 })?;
                             decoded.to_rgb8()
                         };
-                        crate::ocr::layout_assembly::recognize_page_tables(&rgb, det, elements, model)
+                        crate::ocr::layout_assembly::recognize_page_tables(&rgb, scaled_det, elements, model)
                     }
                     _ => Vec::new(),
                 };
@@ -801,8 +827,8 @@ pub(crate) async fn extract_with_ocr(
                 crate::pdf::markdown::reorder_elements_reading_order(&mut page_content.elements);
                 let mut paragraphs = crate::pdf::markdown::content_to_paragraphs(&page_content);
 
-                if let Some(det) = detection {
-                    let hints = detection_to_layout_hints(det, height as f32);
+                if let Some(ref scaled_det) = scaled_detection {
+                    let hints = detection_to_layout_hints(scaled_det, height as f32);
                     crate::pdf::markdown::layout_classify::apply_layout_overrides(
                         &mut paragraphs,
                         &hints,

data/vendor/kreuzberg/src/pdf/error.rs

@@ -76,6 +76,24 @@ pub(crate) fn format_pdfium_error<E: std::fmt::Debug>(error: E) -> String {
     }
 }
 
+/// Classify a PDFium document-load error into the appropriate `PdfError` variant.
+///
+/// Password-related errors are distinguished by inspecting the formatted message.
+/// When `password` is `Some`, a password-related failure means the password was wrong;
+/// when `None`, it means the document requires one.
+pub(crate) fn classify_pdfium_load_error<E: std::fmt::Debug>(error: E, password: Option<&str>) -> PdfError {
+    let err_msg = format_pdfium_error(error);
+    if err_msg.contains("password") || err_msg.contains("Password") {
+        if password.is_some() {
+            PdfError::InvalidPassword
+        } else {
+            PdfError::PasswordRequired
+        }
+    } else {
+        PdfError::InvalidPdf(err_msg)
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;

data/vendor/kreuzberg/src/pdf/mod.rs

@@ -99,7 +99,7 @@ pub use images::{PdfImage, PdfImageExtractor, extract_images_from_pdf};
 #[cfg(feature = "pdf")]
 pub use metadata::extract_metadata;
 #[cfg(feature = "pdf")]
-pub use rendering::{PageRenderOptions, render_page_to_image};
+pub use rendering::{PageRenderOptions, PdfPageIterator, render_page_to_image, render_pdf_page_to_png};
 #[cfg(feature = "pdf")]
 pub use table::extract_words_from_page;
 #[cfg(feature = "pdf")]