Update app.py

app.py

@@ -1,49 +1,59 @@
+import time
+print("Pre-loading models...")
+start = time.time()
+ocr_pipe = pipeline("image-to-text", model="facebook/nougat-base")
+similarity_pipe = pipeline("feature-extraction", model="sentence-transformers/paraphrase-albert-small-v2")
+print(f"Models loaded in {time.time()-start:.2f}s")
 import gradio as gr
-from transformers import pipeline
 import numpy as np
+from PIL import Image
+from transformers import pipeline
 
-# Load FREE models
-ocr_pipe = pipeline("image-to-text", model="microsoft/trocr-base-handwritten")
-similarity_pipe = pipeline("feature-extraction", model="sentence-transformers/all-MiniLM-L6-v2")
+# Load 100% open models (no Microsoft)
+ocr_pipe = pipeline("image-to-text", model="facebook/nougat-base")  # Best for academic handwriting
+# Alternative OCR models: "mfrashad/arabic-handwriting-ocr", "TesseractOCR"
+
+similarity_pipe = pipeline("feature-extraction", model="sentence-transformers/paraphrase-albert-small-v2")  # Lightweight
 
 def validate_answer(image, user_text, correct_answer):
-    # OCR for handwritten text
-    if image:
-        ocr_result = ocr_pipe(image)
-        user_text = ocr_result[0]['generated_text']
-    
-    # Check clarity (rule-based)
-    clarity = sum(c.isalnum() for c in user_text) / max(1, len(user_text))
-    if clarity < 0.7:
-        return "⚠️ Handwriting unclear", "", ""
+    outputs = ["", "", ""]  # Initialize outputs
     
-    # Semantic comparison
-    embeddings = similarity_pipe([correct_answer, user_text])
-    similarity = np.dot(embeddings[0], embeddings[1])
-    
-    return (
-        f"✅ Clarity: {clarity:.0%}",
-        f"📝 Extracted: {user_text}",
-        f"🔍 Similarity: {similarity:.0%}"
-    )
+    try:
+        # OCR Processing
+        if image:
+            img = Image.fromarray(image.astype('uint8'))
+            ocr_result = ocr_pipe(img)
+            user_text = ocr_result[0]['generated_text']
+            outputs[1] = f"📝 Extracted: {user_text}"
+        
+        # Clarity Check (simple rule-based)
+        clarity = sum(c.isalnum() for c in user_text) / max(1, len(user_text))
+        outputs[0] = f"✅ Clarity: {clarity:.0%}" if clarity > 0.5 else "⚠️ Unclear handwriting"
+        
+        # Only compare if text is clear
+        if clarity > 0.5:
+            embeds = np.array(similarity_pipe([correct_answer, user_text]))
+            similarity = np.dot(embeds[0].mean(axis=0), embeds[1].mean(axis=0))
+            outputs[2] = f"🔍 Similarity: {similarity:.1%}"
+            
+    except Exception as e:
+        outputs = [f"❌ Error: {str(e)}"] * 3
+        
+    return outputs
 
-# Create interface
+# Simple Interface
 with gr.Blocks() as demo:
-    gr.Markdown("# Free Answer Validator")
+    gr.Markdown("## Free Handwriting Validator")
     with gr.Row():
-        image_input = gr.Image(label="Upload Handwritten Answer", type="pil")
-        text_input = gr.Textbox(label="Or Type Answer Here")
-    correct_input = gr.Textbox(label="Correct Answer", value="The Earth revolves around the Sun.")
-    submit_btn = gr.Button("Validate")
-    
-    clarity_out = gr.Textbox(label="Clarity Check")
-    extracted_out = gr.Textbox(label="Extracted Text")
-    similarity_out = gr.Textbox(label="Similarity Score")
-    
-    submit_btn.click(
+        gr.Image(label="Upload Answer", sources=["upload"], type="numpy")
+        gr.Textbox(label="Or Type Answer")
+    gr.Textbox(label="Correct Answer", value="Photosynthesis occurs in chloroplasts.")
+    gr.Button("Validate").click(
         validate_answer,
-        inputs=[image_input, text_input, correct_input],
-        outputs=[clarity_out, extracted_out, similarity_out]
+        inputs=[gr.Image(), gr.Textbox(), gr.Textbox()],
+        outputs=[gr.Textbox(label="Status"), 
+                gr.Textbox(label="OCR Result"), 
+                gr.Textbox(label="Comparison")]
     )
 
-demo.launch()
+demo.launch(debug=True)