Skip to main content

Text Detection and Recognition - PaddleOCR

This example runs the PaddleOCR model using the hbm_runtime inference engine for text detection and recognition, supporting OCR recognition and visualization in Chinese scenarios. The example code is located in the /app/pydev_demo/08_OCR_sample/01_paddleOCR/ directory. The example code is also located in the /app/pydev_demo/02_detection_sample/02_ultralytics_yolo11/ directory.

Model Description​

  • Overview:

    This example implements a two-stage OCR task for Chinese text detection and recognition based on PaddleOCR v3. The overall pipeline includes detecting text regions (using a detection model) and recognizing text content region by region (using a recognition model).

  • HBM Model Names:

    • Detection Model: cn_PP-OCRv3_det_infer-deploy_640x640_nv12.hbm
    • Recognition Model: cn_PP-OCRv3_rec_infer-deploy_48x320_rgb.hbm

  • Input Format:

    • Detection Model: BGR image → resized to 640×640 and converted to NV12 format (separated Y and UV planes)
    • Recognition Model: Rotated and cropped BGR text patch → resized to 48×320, normalized, and converted to RGB format

  • Output:

    • Detection Model: Segmentation probability map (1×1×H×W); post-processing yields text bounding box coordinates
    • Recognition Model: Character token logits; decoded via CTC to produce recognized text strings

  • Model Download URLs (automatically downloaded by the program):

    # Detection model
    https://archive.d-robotics.cc/downloads/rdk_model_zoo/rdk_s100/paddle_ocr/cn_PP-OCRv3_det_infer-deploy_640x640_nv12.hbm
    # Recognition model
    https://archive.d-robotics.cc/downloads/rdk_model_zoo/rdk_s100/paddle_ocr/cn_PP-OCRv3_rec_infer-deploy_48x320_rgb.hbm

Functionality Description​

  • Model Loading

    Uses hbm_runtime to load the text detection and recognition models separately, parses input/output names and shapes, and supports setting inference priority and BPU core binding.

  • Input Preprocessing

    • Detection Model: Resizes the original image to 640×640 and converts it to NV12 format (for BPU inference).
    • Recognition Model: Resizes each rotated and cropped text patch to 48×320, converts to RGB format, normalizes the pixel values, and reshapes to NCHW layout.

  • Inference Execution

    Calls the .run() method to perform forward inference, producing a probability map (detection) and logits (recognition).

  • Post-processing

    • Detection Model:

      • Binarizes the probability map using a specified threshold
      • Finds contours of text regions and dilates them
      • Extracts rotated bounding boxes and crops corresponding image regions

    • Recognition Model:

      • Decodes logits using CTCLabelDecode to map them to text strings

    Finally, the recognized text is annotated in red on a blank canvas and stitched with the original image for visualization.

Environment Dependencies​

  • Ensure the dependencies in pydev are installed: 
    pip install -r ../../requirements.txt
  • Install packages required for OCR processing: 
    pip install pyclipper==1.3.0.post6 Pillow==9.0.1 paddlepaddle

Directory Structure​

.
├── FangSong.ttf                # Font for Chinese character display
├── paddle_ocr.py               # Main program implementing text detection and recognition
├── postprocess/                # Post-processing logic (sorting, merging, decoding, etc.)
└── README.md                   # Usage instructions

Parameter Description​

ParameterDefault ValueDescription
--det-model-path/opt/hobot/model/s100/basic/cn_PP-OCRv3_det_infer-deploy_640x640_nv12.hbmPath to the text detection model
--rec-model-path/opt/hobot/model/s100/basic/cn_PP-OCRv3_rec_infer-deploy_48x320_rgb.hbmPath to the text recognition model
--priority0Inference priority (higher value = higher priority)
--bpu-cores[0]Indices of BPU cores to run inference on
--test-img/app/res/assets/gt_2322.jpgInput image path
--label-file/app/res/labels/ppocr_keys_v1.txtPath to label file required for text recognition
--img-save-pathresult.jpgPath to save the inference result image
--threshold0.5Threshold for binarizing text regions
--ratio-prime2.7Expansion factor for text bounding boxes

Quick Start​

  • Run the model

    • With default parameters: 
      python paddle_ocr.py
    • With custom parameters: 
      python paddle_ocr.py \
      --det-model-path /opt/hobot/model/s100/basic/cn_PP-OCRv3_det_infer-deploy_640x640_nv12.hbm \
      --rec-model-path /opt/hobot/model/s100/basic/cn_PP-OCRv3_rec_infer-deploy_48x320_rgb.hbm \
      --test-img /app/res/assets/gt_2322.jpg \
      --label-file /app/res/labels/ppocr_keys_v1.txt \
      --img-save-path result.jpg \
      --priority 0 \
      --bpu-cores 0 \
      --threshold 0.5 \
      --ratio-prime 2.7

  • View Results

    Upon successful execution, the results will be overlaid on the original image and saved to the path specified by --img-save-path:

    [Saved] Result saved to: result.jpg

Notes​

  • If the specified model path does not exist, the program will attempt to download the model automatically.
Last updated on