rafia-10/OCR

Handwritten Amharic OCR using CRNN End-to-end system to recognize handwritten Amharic characters from images using a CNN + LSTM architecture with CTC loss. Got you! Here's a clean, GitHub-friendly version of your README that keeps all the key info but is more concise and scannable: --- # Handwritten Amharic OCR with CRNN A **handwritten Amharic text recognition** system using a **CNN + BiLSTM architecture** with **CTC loss**. Reads cropped images of handwritten Amharic characters and sequences. ## Features * Preprocesses images: grayscale conversion, height normalization, and pixel normalization * **CRNN** for variable-length sequence recognition * Trained with **CTC loss** (alignment-free sequence prediction) * Supports train/validation datasets ## Dataset The dataset is **not included** due to size. Contact for access: * rafiakedir22@gmail.com * danliliyah5@gmail.com ## Usage 1. Clone the repo: ```bash git clone cd ``` 2. Install dependencies: ```bash pip install torch torchvision pillow ``` 3. Prepare dataset structure: ``` Formatted_Data/ ├─ train/ │ ├─ cropped_images/ │ └─ labels/ └─ val/ ├─ cropped_images/ └─ labels/ ``` 4. Run training: ```bash python train.py ``` > Adjust `batch_size` and `learning_rate` as needed. Training time depends on dataset size and GPU availability. ## Notes * Load `amharic_mapping.py` before training * Training may be slow for large datasets; reduce batch size if needed ## Contact For questions, dataset access, or collaboration, email: * rafiakedir22@gmail.com * danliliyah5@gmail.com