example.py

#!/usr/bin/env python3
# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

import argparse

from nemotron_ocr.inference.pipeline import NemotronOCR


def main(image_path, merge_level, no_visualize, model_dir):
    ocr_pipeline = NemotronOCR()

    predictions = ocr_pipeline(image_path, merge_level=merge_level, visualize=not no_visualize)

    print(f"Found {len(predictions)} text regions.")


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="Run OCR inference and annotate image.")
    parser.add_argument("image_path", type=str, help="Path to the input image.")
    parser.add_argument(
        "--merge-level",
        type=str,
        choices=["word", "sentence", "paragraph"],
        default="paragraph",
        help="Merge level for OCR output (word, sentence, paragraph).",
    )
    parser.add_argument("--no-visualize", action="store_true", help="Do not save the annotated image.")
    parser.add_argument(
        "--model-dir",
        type=str,
        help="Path to the model checkpoints.",
        default="./checkpoints",
    )
    args = parser.parse_args()

    main(
        args.image_path,
        merge_level=args.merge_level,
        no_visualize=args.no_visualize,
        model_dir=args.model_dir,
    )