Module layout 

Layout detection via the RT-DETR (docling-layout-heron) model exported to ONNX, run with ort. A port of docling-ibm-models’ LayoutPredictor: resize the page image to 640×640 and rescale to [0,1] (the heron processor has do_normalize=false), run the model, then RT-DETR post_process_object_detection (sigmoid → top-k over query×class → center-to-corners boxes scaled to the page).

Structs

LayoutModel

Region

One detected region, in page points (top-left origin).

Constants

LABELS

The 17 canonical layout classes, indexed by the model’s class id (config.json id2label).

SIDE

RT-DETR’s fixed square input side.

Functions

decode_layout

Decode one page’s raw RT-DETR outputs into scored Regions in page points — sigmoid over every (query, class), top-num_queries kept, boxes converted center→corners and scaled. Shared with the browser build; the native batch path calls it per page, so both decode identically.

label_threshold

Per-label confidence threshold, ported from docling’s LayoutPostprocessor.CONFIDENCE_THRESHOLDS. The raw predictor keeps every detection above the 0.3 base; the postprocessor then drops a cluster whose score is below its label’s threshold. Applying it here (equivalent, since every per-label threshold is ≥ the 0.3 base) keeps low-confidence pictures / tables / list-items out of the assembly, matching docling.

layout_input

Pack one page image into the model’s (1, 3, SIDE, SIDE) input: resize (aspect ignored, RT-DETR convention), rescale to [0,1], CHW. Shared with the browser build (#157), which delegates only the session call.