Module processors 

Modules

db_postprocess

Post-processing for DB (Differentiable Binarization) text detection models.

formula_preprocess

Formula preprocessing utilities for formula recognition models.

layout_postprocess

Layout Detection Post-processing

layout_sorting

Enhanced layout sorting logic — xycut_enhanced algorithm.

layout_utils

Layout parsing utilities.

resize_detection

Image resizing utilities for OCR preprocessing

resize_recognition

OCR-specific image resizing functionality.

simd

SIMD-accelerated kernels for the pipeline’s two hot CPU loops: per-pixel image normalization and per-timestep CTC argmax decode. (Conv/matmul work lives in ONNX Runtime.)

table_ocr_split

Table OCR box splitting utilities.

table_structure_decode

Table Structure Decoding Processor

types

Types used in image processing operations

unimernet_preprocess

UniMERNet-specific preprocessing functions

uvdoc_postprocess

Document transformation post-processing functionality.

Structs

AspectRatioBucket

Aspect ratio bucket definition

AspectRatioBucketing

Aspect ratio bucketing processor

AspectRatioBucketingConfig

Configuration for aspect ratio bucketing

BaseRecLabelDecode

A base decoder for text recognition that handles character mapping and basic decoding operations.

BoundingBox

A bounding box represented by a collection of points.

CTCLabelDecode

A decoder for CTC (Connectionist Temporal Classification) based text recognition models.

CellGridInfo

Grid position and span information for a table cell.

Crop

A processor for cropping images with different positioning modes.

CrossCellDetection

Detection result for a cross-cell OCR box.

DBPostProcess

Post-processor for DB (Differentiable Binarization) text detection models.

DBPostProcessConfig

Runtime configuration for DB post-processing.

DetResizeForTest

A struct for resizing images for OCR testing

FormulaPreprocessParams

Configuration parameters for formula preprocessing pipeline.

FormulaPreprocessor

Formula recognition preprocessor: margin crop → resize-and-pad → normalize-and-grayscale → 4D tensor formatting.

ImageScaleInfo

Information about image scaling during preprocessing

LayoutBox

Layout box with bounding box and label for sorting/processing purposes.

LayoutOCRAssociation

Result of associating OCR boxes with layout regions.

LayoutPostProcess

Layout detection post-processor for models like PicoDet and RT-DETR.

MinAreaRect

A rectangle with minimum area that encloses a shape.

NormalizeImage

Normalizes images for OCR processing.

OCRResize

OCR-specific image resizer.

OverlapRemovalResult

Result of overlap removal.

Point

A 2D point with floating-point coordinates.

SortableRegion

Represents a sortable region for hierarchical sorting.

SplitConfig

Configuration for OCR box splitting.

SplitOcrResult

Result of splitting a single OCR box.

SplitSegment

A segment resulting from splitting an OCR box.

TableStructureDecode

Table structure decoder that converts model outputs to HTML tokens and bboxes.

TableStructureDecodeOutput

Output from table structure decoding.

Topk

A processor for extracting top-k results from classification outputs.

TopkResult

Result structure for top-k classification processing.

UVDocPostProcess

Post-processor for document transformation results.

UniMERNetPreprocessParams

Parameters for UniMERNet preprocessing

UniMERNetPreprocessor

UniMERNet preprocessor that matches Python’s UniMERNetImgDecode logic

Enums

BoxType

Specifies the type of bounding box used for text detection

ColorOrder

Specifies whether the model expects RGB or BGR channel ordering.

CropMode

Specifies how to crop an image when the aspect ratios don’t match

LimitType

Specifies how to limit the size of an image during resizing operations

ResizeType

Specifies different strategies for resizing images

ScoreMode

Specifies the mode for calculating scores in text detection/recognition

SortDirection

Direction for XY-cut sorting.

TensorLayout

Specifies the data layout of an image tensor

Functions

apply_nms_with_merge

Apply Non-Maximum Suppression with per-class merge modes.

assign_elements_to_regions

Assigns elements to regions based on overlap threshold.

associate_ocr_with_layout

Associate OCR results with layout regions.

calculate_iou

Calculates the IoU (Intersection over Union) between two bounding boxes.

calculate_overlap_ratio

Calculates the overlap ratio of box a covered by box b. Returns intersection_area / area_a

create_expanded_ocr_for_table

Creates expanded OCR results after splitting cross-cell boxes.

detect_cross_cell_ocr_boxes

Detects OCR boxes that span multiple table cells.

get_overlap_boxes_idx

Get indices of OCR boxes that overlap with layout regions.

get_overlap_removal_indices

Removes overlapping layout blocks, returning only indices to remove.

merge_boxes

Merge two bounding boxes according to the specified mode.

normalize_latex

Normalizes decoded LaTeX text to match standard output format.

parse_cell_grid_info

Parses structure tokens to extract grid position and span info for each cell.

reconcile_table_cells

Reconciles structure recognition cells with detected cells.

remove_overlap_blocks

Removes overlapping layout blocks based on overlap ratio threshold.

reprocess_table_cells_with_ocr

Reprocesses detected table cell boxes using OCR boxes to better match the structure model’s expected cell count.

sort_boxes_xycut

Sort bounding boxes using XY-cut and return sorted boxes (convenience wrapper).

sort_by_xycut

Sort bounding boxes using the XY-cut algorithm.

sort_elements_with_regions

Sorts elements within each region using XY-cut, then orders regions.

sort_layout_boxes

Sort layout boxes in reading order with column detection.

sort_poly_boxes

Sort polygon boxes (N-point bounding boxes) in reading order.

sort_quad_boxes

Sort quad boxes (4-point bounding boxes) in reading order.

sort_regions

Sorts regions in reading order (top-to-bottom, left-to-right).

sort_with_region_hierarchy

Convenience function that combines region detection output with layout elements to produce a hierarchically sorted reading order.

split_ocr_box_at_cell_boundaries

Splits an OCR box at cell boundaries and distributes text proportionally.

split_text_by_ratio

Splits text into parts based on given ratios.

unclip_boxes

Apply unclip ratio to expand/shrink bounding boxes while keeping center fixed.

wrap_table_html

Wraps table structure tokens with HTML document tags.

wrap_table_html_with_content

Wraps table structure tokens into an HTML string with cell content filled in.

Type Aliases

PositionedDecodeResult

Decoded batch outputs along with positional metadata.