Update app.py

app.py

@@ -1,72 +1,76 @@
 import gradio as gr
+import os
 from PIL import Image
 import numpy as np
 from ultralytics import YOLO
 from openai import OpenAI
-import os
 
+# 🔧 Suppress Ultralytics config warning in Spaces
+os.environ["YOLO_CONFIG_DIR"] = "/tmp"
+
+# ✅ Initialize Groq client with Hugging Face secret
 client = OpenAI(
-    api_key=os.getenv("civil_project"),  # Hugging Face reads from environment variables
+    api_key=os.getenv("civil_project"),
     base_url="https://api.groq.com/openai/v1"
 )
-# ✅ Load YOLOv8 model
-model = YOLO("yolov8n.pt")  # You can later swap with a custom-trained model
 
-# ✅ Function to ask Groq LLaMA 3 model for repair advice
+# ✅ Load crack-segmentation model from Hugging Face
+# Use 'yolov8n-seg.pt' for fast edge inference
+model = YOLO("yolov8n-seg.pt")  # Downloaded from 'OpenSistemas/YOLOv8-crack-seg' :contentReference[oaicite:1]{index=1}
+
 def ask_groq(prompt):
-    response = client.chat.completions.create(
-        model="llama3-8b-8192",  # Groq-supported model
+    resp = client.chat.completions.create(
+        model="llama3-8b-8192",
         messages=[
-            {"role": "system", "content": "You are an expert in construction damage assessment."},
+            {"role": "system", "content": "You are an expert structural engineer specialized in crack diagnosis."},
             {"role": "user", "content": prompt}
         ],
         temperature=0.5
     )
-    return response.choices[0].message.content.strip()
+    return resp.choices[0].message.content.strip()
 
-# ✅ Damage detection + chatbot response
 def process_image(image):
     try:
-        # Convert image from NumPy to PIL
         if isinstance(image, np.ndarray):
             image = Image.fromarray(image)
         elif isinstance(image, str):
             image = Image.open(image)
 
-        # Run YOLOv8 for object detection
-        results = model(image)
-        labels = results[0].names
-        boxes = results[0].boxes
-        detected = [labels[int(cls)] for cls in boxes.cls]
+        # 🔍 Crack segmentation inference
+        results = model.predict(source=image, imgsz=640, verbose=False)[0]
+        masks = results.masks.data if results.masks is not None else []
+        count = len(masks)
 
-        detected_info = ", ".join(set(detected)) or "no damage detected"
+        if count > 0:
+            detected_info = f"{count} crack(s) detected"
+            prompt_prefix = detected_info
+        else:
+            prompt_prefix = (
+                "No cracks were detected by the vision model, "
+                "but the image may contain hidden surface damage. "
+                "Please analyze contextually."
+            )
 
-        # Prompt sent to Groq's LLaMA 3 model
         user_prompt = f"""
-        The uploaded image shows: {detected_info}.
+        {prompt_prefix}.
         Please:
-        - Diagnose the construction issue
-        - Suggest possible repair solutions
+        - Diagnose the issue
+        - Suggest repair methods
         - List tools/materials required
         - Estimate repair time
         """
 
-        response = ask_groq(user_prompt)
-        return response
+        return ask_groq(user_prompt)
 
     except Exception as e:
-        return f"❌ Error: {str(e)}"
+        return f"❌ Error: {e}"
 
-# ✅ Gradio UI
 with gr.Blocks() as demo:
-    gr.Markdown("## 🏗️ Construction Damage Analyzer")
-    gr.Markdown("Upload a photo of damaged wall, pipe, or structure. Get diagnosis and repair suggestions from AI.")
-
-    with gr.Row():
-        img_input = gr.Image(type="numpy", label="Upload Damage Image")
-        output_text = gr.Textbox(label="Diagnosis & Suggestions")
+    gr.Markdown("## 🚧 Construction Crack Analyzer")
+    gr.Markdown("Upload an image of a wall or surface to detect and analyze cracks.")
 
-    submit_btn = gr.Button("Analyze")
-    submit_btn.click(fn=process_image, inputs=img_input, outputs=output_text)
+    img_input = gr.Image(type="numpy", label="Upload Damage Image")
+    output_text = gr.Textbox(label="Diagnosis & Recommendations", lines=8)
+    gr.Button("Analyze").click(fn=process_image, inputs=img_input, outputs=output_text)
 
 demo.launch()