Struct StructureResult 

pub struct StructureResult {
    pub input_path: Arc<str>,
    pub index: usize,
    pub layout_elements: Vec<LayoutElement>,
    pub tables: Vec<TableResult>,
    pub formulas: Vec<FormulaResult>,
    pub text_regions: Option<Vec<TextRegion>>,
    pub orientation_angle: Option<f32>,
    pub region_blocks: Option<Vec<RegionBlock>>,
    pub rectified_img: Option<Arc<ImageBuffer<Rgb<u8>, Vec<u8>>>>,
    pub page_continuation_flags: Option<PageContinuationFlags>,
}

Result of document structure analysis.

This struct contains all the results from analyzing a document’s structure, including layout elements, tables, formulas, and OCR results.

Coordinate System

The coordinate system of bounding boxes depends on which preprocessing was applied:

• No preprocessing: Boxes are in the original input image’s coordinate system.

• Orientation correction only (orientation_angle set, rectified_img is None): Boxes are transformed back to the original input image’s coordinate system.

• Rectification applied (rectified_img is Some): Boxes remain in the rectified image’s coordinate system. Neural network-based rectification (UVDoc) warps cannot be precisely inverted, so use rectified_img for visualization instead of the original image.

• Both orientation and rectification: Boxes are in the rectified coordinate system (rectification takes precedence since it’s applied after orientation correction).

Fields

input_path: Arc<str>

Path to the input image file

index: usize

Index of the image in a batch (0 for single image processing)

layout_elements: Vec<LayoutElement>

Detected layout elements (text regions, tables, figures, etc.)

tables: Vec<TableResult>

Recognized tables with their structure and content

formulas: Vec<FormulaResult>

Recognized mathematical formulas

text_regions: Option<Vec<TextRegion>>

OCR text regions (if OCR was integrated)

orientation_angle: Option<f32>

Document orientation angle (if orientation correction was used)

region_blocks: Option<Vec<RegionBlock>>

Detected region blocks for hierarchical ordering (PP-DocBlockLayout) When present, layout_elements are already sorted by region hierarchy

rectified_img: Option<Arc<ImageBuffer<Rgb<u8>, Vec<u8>>>>

Rectified image (if document rectification was used) Note: Bounding boxes are already transformed back to original coordinates for rotation, but for rectification (UVDoc), boxes are in the rectified image’s coordinate system. Use this image for visualization when rectification was applied.

page_continuation_flags: Option<PageContinuationFlags>

Page continuation flags for multi-page document processing. This indicates whether this page continues a paragraph from the previous page or continues to the next page, which is crucial for proper markdown concatenation.

Implementations

impl StructureResult

pub fn new(input_path: impl Into<Arc<str>>, index: usize) -> StructureResult

Creates a new structure result.

pub fn with_layout_elements( self, elements: Vec<LayoutElement>, ) -> StructureResult

Adds layout elements to the result.

pub fn with_tables(self, tables: Vec<TableResult>) -> StructureResult

Adds tables to the result.

pub fn with_formulas(self, formulas: Vec<FormulaResult>) -> StructureResult

Adds formulas to the result.

pub fn with_text_regions(self, regions: Vec<TextRegion>) -> StructureResult

Adds OCR text regions to the result.

pub fn with_region_blocks(self, blocks: Vec<RegionBlock>) -> StructureResult

Adds region blocks to the result (PP-DocBlockLayout).

Region blocks represent hierarchical groupings of layout elements. When set, layout_elements should already be sorted by region hierarchy.

pub fn with_page_continuation_flags( self, flags: PageContinuationFlags, ) -> StructureResult

Sets page continuation flags for multi-page document processing.

pub fn to_markdown(&self) -> String

Converts the result to a Markdown string.

Follows PP-StructureV3’s formatting rules:

• DocTitle: # title
• ParagraphTitle: Auto-detect numbering (1.2.3 -> ###)
• Formula: $$latex$$
• Table: HTML with border
• Images: ![Figure](caption)

Note: Low-confidence text elements that overlap with table regions are filtered out to avoid duplicate content from table OCR.

pub fn calculate_continuation_flags(&self) -> PageContinuationFlags

Calculates the page continuation flags for this result.

This follows PaddleX’s get_seg_flag logic to determine whether the page starts/ends in the middle of a semantic paragraph.

Returns (paragraph_start, paragraph_end) where:

• paragraph_start: false means page continues from previous
• paragraph_end: false means content continues to next page

pub fn to_html(&self) -> String

Converts the result to an HTML string.

Follows PP-StructureV3’s formatting rules with semantic HTML tags.

pub fn to_json_value(&self) -> Result<Value, Error>

Converts the result to a JSON Value.

pub fn save_results( &self, output_dir: impl AsRef<Path>, to_json: bool, to_html: bool, ) -> Result<(), Error>

Saves the analysis results to the specified directory.

This generates:

• *_res.json: The full structured result
• *_res.html: An HTML representation

Note: Markdown export with image extraction should use the example utilities (examples/utils/markdown.rs) instead, as that requires I/O operations that belong in the application layer. Use StructureResult::to_markdown() for pure markdown generation without side effects.

Arguments

• output_dir - Directory to save the output files
• to_json - If true, save a JSON representation
• to_html - If true, save an HTML representation

Trait Implementations

impl Clone for StructureResult

fn clone(&self) -> StructureResult

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for StructureResult

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for StructureResult

fn deserialize<__D>( __deserializer: __D, ) -> Result<StructureResult, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for StructureResult

fn serialize<__S>( &self, __serializer: __S, ) -> Result<<__S as Serializer>::Ok, <__S as Serializer>::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructureResultExt for StructureResult

fn to_concatenated_markdown(results: &[StructureResult]) -> String

Converts multiple results to a single concatenated markdown.

fn save_multi_page_results( results: &[StructureResult], output_dir: impl AsRef<Path>, base_name: &str, to_json: bool, to_markdown: bool, to_html: bool, ) -> Result<(), Error>

where StructureResult: Sized,

Saves multiple results with concatenated markdown.