Upload README.md with huggingface_hub

README.md

@@ -10,13 +10,13 @@ tags:
 - pytorch
 ---
 
-# Qalam-Net (قلم-نت): Advanced Arabic OCR
+# Qalam-Net (قلم-نت): Advanced Arabic OCR (v2 Portable)
 
-Qalam-Net is a high-performance, cross-backend Optical Character Recognition (OCR) model for Arabic. Built on **Keras 3**, it supports **JAX**, **PyTorch**, and **TensorFlow** backends.
+Qalam-Net is a high-performance, cross-backend Optical Character Recognition (OCR) model for Arabic. This version (**v2**) has been patched for maximum portability across Keras versions and optimized for inference.
 
-## 🚀 Quick Start (Advanced Usage)
+## 🚀 Quick Start (Clean Inference)
 
-To use Qalam-Net, you must define the `CTCLayer` (used during training) and pass it as a custom object.
+Qalam-Net v2 no longer requires custom layers or complex setup during inference.
 
 ### 1. Installation
 ```bash
@@ -26,69 +26,48 @@ pip install -U "keras>=3.0" jax jaxlib huggingface_hub opencv-python
 ### 2. Implementation
 ```python
 import os
-os.environ["KERAS_BACKEND"] = "jax" # Switch to "tensorflow" or "torch" if preferred
+os.environ["KERAS_BACKEND"] = "jax" # Options: "jax", "tensorflow", "torch"
 
 import keras
-from keras import layers
 import numpy as np
 import cv2
-from huggingface_hub import hf_hub_download
-
-# [MANDATORY] Define the CTCLayer for deserialization
-@keras.saving.register_keras_serializable()
-class CTCLayer(layers.Layer):
-    def __init__(self, name=None, **kwargs):
-        super().__init__(name=name, **kwargs)
-        self.loss_fn = keras.backend.ctc_batch_cost
-
-    def call(self, y_true, y_pred):
-        return y_pred
 
 class QalamNet:
     def __init__(self, repo_id="Ali0044/Qalam-Net"):
-        # Download the model
-        model_path = hf_hub_download(repo_id=repo_id, filename="model.keras")
-        
-        # Load with custom_objects
-        self.model = keras.saving.load_model(
-            model_path, 
-            custom_objects={"CTCLayer": CTCLayer},
-            compile=False
-        )
+        # Load the portable model directly from Hugging Face
+        # No 'custom_objects' or 'CTCLayer' required!
+        self.model = keras.saving.load_model(f"hf://{repo_id}")
         
         # Standard Arabic Vocabulary
         self.vocab = [' ', '!', '"', '#', '(', ')', '*', '+', ',', '-', '.', '/', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', ':', ';', '=', '?', '[', ']', 'ء', 'آ', 'أ', 'ؤ', 'إ', 'ئ', 'ا', 'ب', 'ة', 'ت', 'ث', 'ج', 'ح', 'خ', 'د', 'ذ', 'ر', 'ز', 'س', 'ش', 'ص', 'ض', 'ط', 'ظ', 'ع', 'غ', 'ـ', 'ف', 'ق', 'ك', 'ل', 'م', 'ن', 'ه', 'و', 'ى', 'ي', 'ً', 'ٌ', 'ٍ', 'َ', 'ُ', 'ِ', 'ّ', 'ْ', '٠', '١', '٢', '٣', '٤', '٥', '٦', '٧', '٨', '٩']
         
     def preprocess(self, image_path):
         img = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE)
-        img = cv2.resize(img, (128, 32))
-        img = (img / 255.0).astype(np.float32)
-        img = img.T
-        img = np.expand_dims(img, axis=-1)
-        return np.expand_dims(img, axis=0)
+        img = cv2.resize(img, (128, 32)) / 255.0
+        img = img.T # (Height, Width) -> (Width, Height)
+        img = np.expand_dims(img, axis=(-1, 0)) # Add Channel and Batch
+        return img.astype(np.float32)
 
     def predict(self, image_path):
         batch_img = self.preprocess(image_path)
-        # The model has 2 inputs [image, label] but we only need image for prediction
-        # We can pass dummy labels or use the internal prediction layers
-        preds = self.model.predict([batch_img, np.zeros((1, 1))])
+        predictions = self.model.predict(batch_img)
         
         # CTC Decode
-        input_len = np.ones(preds.shape[0]) * preds.shape[1]
-        results = keras.backend.ctc_decode(preds, input_length=input_len, greedy=True)[0][0]
+        input_len = np.ones(predictions.shape[0]) * predictions.shape[1]
+        results = keras.backend.ctc_decode(predictions, input_length=input_len, greedy=True)[0][0]
         
         return "".join([self.vocab[int(res)] for res in results[0] if res != -1])
 
-# Usage
+# Run
 # ocr = QalamNet()
-# print(ocr.predict("text.png"))
+# print(f"Output: {ocr.predict('test_sample.png')}")
 ```
 
 ## 🧠 Model Architecture
 Qalam-Net employs a specialized **CNN-BiLSTM-Attention** pipeline:
-- **Spatial Features**: 3-block CNN for robust feature extraction.
-- **Sequence Context**: Dual Bidirectional LSTMs to capture Arabic script flow.
-- **Focus Mechanism**: Self-attention layer for character-level precision.
+- **CNN Backbone**: Extracts high-level spatial features from Arabic script.
+- **BiLSTM Layers**: Captures the sequential nature of right-to-left writing.
+- **Attention Mechanism**: Resolves difficult character boundaries.
 
 ---
-**Developed by [Ali Khalid](https://github.com/Ali0044)**
+**Maintained by [Ali Khalid](https://github.com/Ali0044)**