anminhhung / smart_menu_ocr

Description

• Input: Menu images
• Output:
  • Merge food name and price
  • Translate Vietnamese to English

---

Survey

• Text Detection: Pannet, Craft, EAST, DBNet, Bezier.

• Text Recognition: VietOCR, DeepText, Aster, MORAN.

• E2E: PPOCR, mmOCR, AdelaiDet.

• Post_processing: Annoy, RapidFuzz, FuzzyWuzzy, Difflib, Threading, Sort Point.

• Image Rotation: Document-Scanner-and-OCR, Document Scanner, Deskew, skew_correction.

• Translation: Hugging Face, Machine Translation.

---

Problem Solving Strategies

---

Performance

VinAI Dataset

Detection	Recognition	Time with CPU
EAST-ResNet50_vd	SVTR-Tiny	18.0 (s)
EAST-ResNet50_vd	PPOCRV3 (Quantization)	16.0 (s)
EAST-MobileNetV3	SVTR-Tiny	13.0 (s)
EAST-MobileNetV3	PPOCRV3 (Quantization)	12.0 (s)
PPOCRV3 (Convert ONNX)	PPOCRV3 (Convert ONNX)	3.0 (s)

---

Train model

1. Text Detection: Text Detection.
2. Text Recognition: Text Recognition.

Inference (Convert ONNX)

1. Colab
   • Link colab: link.

---

Build and run docker

1. Build docker

   • Create dockerfile: link Dockerfile.

   • Build image:

       docker build -t <name_image>:<version> .
     

2. Run Docker

   • Run image:

       docker run --name <container_name> -p 5000:5000 <image_name>
     

3. Save Docker

   • Save image:

       docker save -o <name>.tar <name_image>