crnn-fa-printed-96-long
| Property | Value |
|---|---|
| License | Apache 2.0 |
| Language | Persian (Farsi) |
| Research Paper | CRNN Paper |
| Downloads | 24,564 |
What is crnn-fa-printed-96-long?
This is an advanced Persian OCR (Optical Character Recognition) model based on the CRNN (Convolutional Recurrent Neural Network) architecture. It represents a significant improvement over its predecessor, specifically designed to handle printed Persian text with enhanced capabilities and broader character support.
Implementation Details
The model combines CNN and LSTM architectures, optimized for processing Persian text images. It features a revised input image size of 32x384 pixels and can handle sequences up to 96 characters, though it's optimized for texts around 50 characters in length.
- 5X larger training dataset compared to previous version
- Modified input dimensions (32x384)
- Extended maximum output length to 96 characters
- Intelligent handling of LTR characters within RTL text
- Comprehensive support for numbers and special characters
Core Capabilities
- High-accuracy Persian text recognition from printed documents
- Automatic handling of mixed RTL/LTR text
- Support for numbers and special characters
- Optimized for word-level text detection
- Suitable for fine-tuning on specific domains
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its enhanced capability to handle longer text sequences (up to 96 characters) and its sophisticated handling of mixed RTL/LTR text, making it particularly effective for real-world Persian document processing.
Q: What are the recommended use cases?
The model is primarily designed for printed/scanned documents and works best with text boxes containing up to 50 characters. It's recommended to use it in conjunction with a text detector model for optimal results in end-to-end OCR pipelines.
