Update app.py

app.py

@@ -1,72 +1,69 @@
 import gradio as gr
 import easyocr
+from googletrans import Translator
 from PIL import Image, ImageDraw, ImageFont
 import numpy as np
-from transformers import MarianMTModel, MarianTokenizer
-import io
+import cv2
 
 # Load OCR reader
-reader = easyocr.Reader(['en'])
+reader = easyocr.Reader(['en', 'hi'])
 
-# Translation model - English to Hindi
-model_name = "Helsinki-NLP/opus-mt-en-hi"
-tokenizer = MarianTokenizer.from_pretrained(model_name)
-model = MarianMTModel.from_pretrained(model_name)
+# Translator
+translator = Translator()
 
-# Font path (this works on Hugging Face Spaces)
-FONT_PATH = "/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf"
+def translate_image(image, src_lang, tgt_lang):
+    # Convert PIL Image to OpenCV format
+    img_np = np.array(image)
+    img_cv = cv2.cvtColor(img_np, cv2.COLOR_RGB2BGR)
 
-# Translation function (cleaned and CPU safe)
-def translate_text(text):
-    inputs = tokenizer(text, return_tensors="pt")
-    translated = model.generate(**inputs)
-    return tokenizer.decode(translated[0], skip_special_tokens=True)
+    # OCR
+    results = reader.readtext(img_cv)
 
-# Main function
-def process_image(image, font_size):
-    np_image = np.array(image)
-    results = reader.readtext(np_image)
-    
     if not results:
-        return image, "No text found."
+        return "No text detected.", image
 
-    # Convert image to editable mode
-    img_pil = image.convert("RGB")
-    draw = ImageDraw.Draw(img_pil)
-    font = ImageFont.truetype(FONT_PATH, size=font_size)
-    
-    all_translations = []
+    draw = ImageDraw.Draw(image)
+    font = ImageFont.truetype("arial.ttf", 22)
 
     for (bbox, text, prob) in results:
-        if text.strip():
-            try:
-                translated = translate_text(text)
-            except:
-                translated = "[Translation Error]"
+        # Translate text
+        try:
+            translated = translator.translate(text, src=src_lang, dest=tgt_lang).text
+        except:
+            translated = text  # fallback
 
-            all_translations.append(translated)
+        # Coordinates
+        top_left = tuple(map(int, bbox[0]))
+        bottom_right = tuple(map(int, bbox[2]))
 
-            # Position text above the bounding box
-            x, y = bbox[0]
-            draw.text((x, y - font_size), translated, fill=(255, 0, 0), font=font)
+        # Draw rectangle over original text (optional blur it)
+        draw.rectangle([top_left, bottom_right], fill="white")
 
-    full_text = "\n".join(all_translations)
-    return img_pil, full_text
+        # Write translated text
+        draw.text(top_left, translated, fill="black", font=font)
+
+    return image
+
+# Language options for dropdown (EasyOCR short codes → Googletrans short codes)
+lang_map = {
+    "English": "en",
+    "Hindi": "hi",
+    "French": "fr",
+    "Spanish": "es",
+    "German": "de",
+    "Chinese": "zh-cn"
+}
 
-# Gradio interface
 iface = gr.Interface(
-    fn=process_image,
+    fn=translate_image,
     inputs=[
         gr.Image(type="pil", label="Upload Image"),
-        gr.Slider(20, 80, value=30, label="Overlay Font Size")
-    ],
-    outputs=[
-        gr.Image(type="pil", label="Image with Translated Overlay"),
-        gr.Textbox(label="Translated Text Output")
+        gr.Dropdown(list(lang_map.keys()), label="From Language", value="Hindi"),
+        gr.Dropdown(list(lang_map.keys()), label="To Language", value="English"),
     ],
-    title="Offline Image Translator",
-    description="Upload an image with English text. It'll translate it to Hindi and overlay the result directly on the image. No Cloudinary used!"
+    outputs=gr.Image(type="pil", label="Translated Image"),
+    title="Photo Translate",
+    description="Upload an image with text. It will detect the text, translate it, and regenerate the image with the translated text."
 )
 
-if __name__ == "__main__":
-    iface.launch(share=True)
+iface.launch()