WjdZddlmZddZdS)aD Image Model Features (IMF) extraction Maintained by: AI Researchers Purpose: Extract image-based features using neural network (ONNX) image_feature_extraction_task() is the stable interface for pymupdf_util_ext.py. Inference and detection logic lives in ImageFeatureExtractorV1 / V2. feature_map / class_logits are returned SEPARATELY (not concatenated): they have different statistical character (continuous embedding vs. per-class score) and are meant to be pooled differently downstream -- see roi_pooling.DEFAULT_FEATURE_MAP_POOLING_OPS / DEFAULT_CLASS_LOGITS_POOLING_OPS. ) BOX_IMAGENcP|s||||}|}g}g}d|vrF|D]1}|||t 2||||fS)u Extract image-based features and segmentation-derived bboxes. Args: page_img: np.ndarray (H, W, C), uint8 — page raster image feature_extractor: ImageFeatureExtractorV1 / V2 instance input_type: tuple of element types; 'seg-image' enables bbox detection aug_fetmap: optional extra channel map passed to predict() Returns: feature_map: decoder embedding output, shape (1, 5*F, H, W) class_logits: per-class segmentation logits, shape (1, C, H, W) bboxes_to_add: list of [x1, y1, x2, y2] in page pixel space box_types_to_add: list of box type strings (parallel to bboxes_to_add) ) aug_fetmapz seg-image) consume_cachepredictget_feature_mapget_class_logitsget_picture_detectionsappendr) page_imgfeature_extractor input_typer feature_map class_logits bboxes_to_addbox_types_to_addbboxs /Users/tedpeters/Documents/Projects/Claude Projects/TPC Houses/house-viewer/.venv/lib/python3.11/site-packages/pymupdf/layout/pymupdf_util_imf.pyimage_feature_extraction_taskrs"  * * , ,C!!(z!BBB#3355K$5577LMj  %<<>> / /D   & & &  # #I . . . .  m5E EE)N)__doc__pymupdf_util_baserrrrrsG  )(((((FFFFFFr