Update app.py

app.py

@@ -3,32 +3,38 @@ import cv2
 import numpy as np
 from PIL import Image
 import pickle
+import tensorflow as tf
 from tensorflow.keras.models import load_model
 from tensorflow.keras.preprocessing.image import img_to_array
 import easyocr
+import torch
 
-# === Load Model and Label Encoder ===
+# ========== GPU Checks ==========
+print("Torch GPU Available:", torch.cuda.is_available())
+print("TensorFlow GPU Devices:", tf.config.list_physical_devices('GPU'))
+
+# ========== Load Model and Label Encoder ==========
 model_path = "MobileNetBest_Model.h5"
 label_path = "MobileNet_Label_Encoder.pkl"
 
 model = load_model(model_path)
-print("Model loaded.")
+print("✅ MobileNet model loaded.")
 
-# Load label encoder
+# Label encoder
 try:
     with open(label_path, 'rb') as f:
         label_map = pickle.load(f)
         index_to_label = {v: k for k, v in label_map.items()}
-    print("Label encoder loaded:", index_to_label)
+    print("✅ Label encoder loaded:", index_to_label)
 except:
     index_to_label = {0: "Handwritten", 1: "Computerized"}
-    print("Label encoder not found. Using default:", index_to_label)
+    print("⚠️ Default labels used:", index_to_label)
 
-# === Initialize EasyOCR Reader Once (with GPU) ===
-reader = easyocr.Reader(['en'], gpu=True)
-print("EasyOCR Reader initialized with GPU.")
+# ========== Initialize EasyOCR (Force GPU) ==========
+reader = easyocr.Reader(['en'], gpu=torch.cuda.is_available())
+print("✅ EasyOCR initialized with GPU:", torch.cuda.is_available())
 
-# === Classify Region ===
+# ========== Classify One Region ==========
 def classify_text_region(region_img):
     try:
         region_img = cv2.resize(region_img, (224, 224))
@@ -44,10 +50,10 @@ def classify_text_region(region_img):
             class_idx = np.argmax(preds[0])
             return index_to_label.get(class_idx, "Unknown")
     except Exception as e:
-        print("Classification error:", e)
+        print("❌ Classification error:", e)
         return "Unknown"
 
-# === OCR + Annotation ===
+# ========== OCR & Annotate ==========
 def AnnotatedTextDetection_EasyOCR_from_array(img):
     results = reader.readtext(img)
     annotated_results = []
@@ -59,6 +65,7 @@ def AnnotatedTextDetection_EasyOCR_from_array(img):
         x1, y1 = map(int, bbox[0])
         x2, y2 = map(int, bbox[2])
         crop = img[y1:y2, x1:x2]
+
         if crop.size == 0:
             continue
 
@@ -71,11 +78,10 @@ def AnnotatedTextDetection_EasyOCR_from_array(img):
 
     return cv2.cvtColor(img, cv2.COLOR_BGR2RGB), "\n".join(annotated_results)
 
-# === Gradio Wrapper ===
+# ========== Inference Function ==========
 def infer(image):
     img = np.array(image)
 
-    # Resize if image is too large
     max_dim = 1000
     if img.shape[0] > max_dim or img.shape[1] > max_dim:
         scale = max_dim / max(img.shape[0], img.shape[1])
@@ -84,7 +90,7 @@ def infer(image):
     annotated_img, result_text = AnnotatedTextDetection_EasyOCR_from_array(img)
     return Image.fromarray(annotated_img), result_text
 
-# === Custom CSS ===
+# ========== Gradio UI ==========
 custom_css = """
 body {
     background-color: #e6f2ff;
@@ -100,7 +106,6 @@ body {
 }
 """
 
-# === Launch Interface ===
 demo = gr.Interface(
     fn=infer,
     inputs=gr.Image(type="pil", label="Upload Image"),
@@ -109,8 +114,9 @@ demo = gr.Interface(
         gr.Textbox(label="Detected Text and Classification")
     ],
     title="Text Detection and Classification",
-    description="This application detects text using EasyOCR and classifies each text region as Handwritten or Computerized using a MobileNet model.",
+    description="Application detects text from images and classify into Handwritten/Computerized Text",
     theme="soft",
     css=custom_css
 )
+
 demo.launch()