Calculate target size with aspect ratio preservation.

Call detection task

Call Doc task (supports image and PDF input)

Call OCR task

Call recognition task

Call UniRec task (supports image and PDF input)

compute the distance from point to a line ys: coordinates in the first axis xs: coordinates in the second axis point_

Internal method which performs the necessary filtering prior to decoding.

Determine execution providers based on GPU availability and user preference. Args: use_gpu: None (auto-detect), True (force GPU),

Determine execution providers based on GPU availability and user preference. Args: use_gpu: None (auto-detect), True (force GPU),

Handle figure content

Convert section headings to appropriate markdown format

Convert list items to markdown list format

Run inference on a single BGR numpy image (used for PDF pages). Args: img_numpy: Input image as numpy array (BGR format)

Initialize detection task

Initialize Doc task (document OCR with layout)

Initialize OCR task (detection + recognition)

Initialize recognition task

Initialize UniRec task (universal recognition)

Load config from file. Args: file_path (str): Path of the config file to be loaded. Returns: global config

Run VLM recognition on multiple blocks in parallel. Uses ThreadPoolExecutor with multiple UniRecONNX instances to process up to max_p

Convert PDF file to a list of BGR numpy images. Args: pdf_path: Path to PDF file Returns: List of numpy arra

Convert PDF file to a list of PIL Images. Args: pdf_path: Path to PDF file Returns: List of PIL Image object

Process mathematical formulas in text by iteratively finding and replacing formulas. - Identify inline and block formulas - R

Recognize a single block using the shared VLM recognizer. The underlying ONNX Runtime session is thread-safe, so multiple threads can

Remove newline in heading

r"""Initiate parameters in the transformer model.

Gasuss noise.

Returns the names of available CLIP models

Calculate the IOU between two boundaries. Args: src (list): Source boundary. target (list): Target boundary. Returns: i

box_score_slow: use polyon mean score as the mean score

_bitmap: single map with shape (1, H, W), whose values are binarized as {0, 1}

convert text-index into text-label.

Return coordinates of fiducial points in I_r; C.

Return inv_delta_C which is needed to calculate T.

Build the wheel package

Args: p: probability distribution over the last dimension, of size (..., L, C) mask: (..., L)