updated

.gitignore

@@ -1 +1,2 @@
 .env
+yolov8x-world.pt.eac99ff4aff54a2a95f4462dc49b3d49.partial

app.py

@@ -2,50 +2,89 @@ import gradio as gr
 from huggingface_hub import InferenceClient
 import cv2
 import numpy as np
-from mcp import MCP
 import time
 import os
 from datetime import datetime
+from ultralytics import YOLO
+from transformers import AutoProcessor, AutoModelForCausalLM
+import torch
 
 """
 For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
 """
 client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
-mcp = MCP()
 
-def detect_fire(frame):
-    # Convert frame to HSV color space
-    hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
-    
-    # Define range for fire colors (red and orange)
-    lower_fire = np.array([0, 50, 50])
-    upper_fire = np.array([30, 255, 255])
+# Load YOLO-World model
+model = YOLO('yolov8x-world.pt')
+
+# Load CLIP model for image understanding
+processor = AutoProcessor.from_pretrained("openai/clip-vit-base-patch32")
+clip_model = AutoModelForCausalLM.from_pretrained("openai/clip-vit-base-patch32")
+
+def analyze_fire_scene(frame):
+    # Run YOLO-World inference with custom prompts
+    results = model(frame, text=["fire", "flame", "smoke", "burning", "wildfire"])
     
-    # Create mask for fire colors
-    mask = cv2.inRange(hsv, lower_fire, upper_fire)
+    # Initialize detection flags and details
+    fire_detected = False
+    smoke_detected = False
+    fire_details = []
     
-    # Calculate percentage of fire-colored pixels
-    fire_percentage = (np.sum(mask > 0) / (frame.shape[0] * frame.shape[1])) * 100
+    # Process results
+    for result in results:
+        boxes = result.boxes
+        for box in boxes:
+            confidence = float(box.conf[0])
+            if confidence > 0.5:
+                class_name = result.names[int(box.cls[0])]
+                if class_name in ['fire', 'flame', 'burning', 'wildfire']:
+                    fire_detected = True
+                    # Get bounding box coordinates
+                    x1, y1, x2, y2 = box.xyxy[0].cpu().numpy()
+                    # Extract the region of interest
+                    roi = frame[int(y1):int(y2), int(x1):int(x2)]
+                    fire_details.append({
+                        'type': class_name,
+                        'confidence': confidence,
+                        'location': (x1, y1, x2, y2),
+                        'roi': roi
+                    })
+                elif class_name == 'smoke':
+                    smoke_detected = True
     
-    return fire_percentage > 5  # Return True if more than 5% of pixels are fire-colored
+    return fire_detected, smoke_detected, fire_details
 
-def send_alert(detection_type, location):
-    # Get current timestamp
-    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
-    message = f"{detection_type} detected at {location} on {timestamp}"
+def get_fire_analysis(frame, fire_details):
+    # Prepare image for CLIP
+    inputs = processor(images=frame, return_tensors="pt")
     
-    # Send SMS using MCP
-    mcp.sms.send(
-        to="YOUR_PHONE_NUMBER",  # Replace with actual phone number
-        message=message
-    )
+    # Generate questions about the fire
+    questions = [
+        "What is the intensity of the fire?",
+        "Is the fire spreading?",
+        "What is the color of the smoke?",
+        "Are there any people or buildings nearby?",
+        "What is the approximate size of the fire?"
+    ]
     
-    # Send email using MCP
-    mcp.email.send(
-        to="YOUR_EMAIL",  # Replace with actual email
-        subject=f"{detection_type} Alert",
-        body=message
-    )
+    analysis = []
+    for question in questions:
+        # Process question with CLIP
+        text_inputs = processor(text=question, return_tensors="pt", padding=True)
+        
+        # Get image-text similarity
+        with torch.no_grad():
+            image_features = clip_model.get_image_features(**inputs)
+            text_features = clip_model.get_text_features(**text_inputs)
+            
+            # Calculate similarity
+            similarity = torch.nn.functional.cosine_similarity(image_features, text_features)
+            
+            # Generate response based on similarity
+            if similarity > 0.5:
+                analysis.append(f"Q: {question}\nA: Based on visual analysis, {question.lower()}")
+    
+    return analysis
 
 def check_for_fire():
     # Request webcam access
@@ -59,19 +98,23 @@ def check_for_fire():
         cap.release()
         return "Error: Could not read from webcam"
     
-    # Detect fire
-    fire_detected = detect_fire(frame)
+    # Detect fire and smoke
+    fire_detected, smoke_detected, fire_details = analyze_fire_scene(frame)
     
     # Release webcam
     cap.release()
     
+    # Get location (you might want to implement a more sophisticated location detection)
+    location = "Webcam Location"  # Replace with actual location detection
+    
     if fire_detected:
-        # Get location (you might want to implement a more sophisticated location detection)
-        location = "Webcam Location"  # Replace with actual location detection
-        send_alert("Fire", location)
-        return f"Fire detected at {location}! Alerts have been sent."
+        # Get detailed analysis of the fire
+        analysis = get_fire_analysis(frame, fire_details)
+        return f"Fire detected at {location}!\n\nAnalysis:\n" + "\n".join(analysis)
+    elif smoke_detected:
+        return f"Smoke detected at {location}!"
     else:
-        return "No fire detected"
+        return "No fire or smoke detected"
 
 def respond(
     message,

pyproject.toml

@@ -3,7 +3,7 @@ name = "wild-fire-tracker"
 version = "0.1.0"
 description = "Add your description here"
 readme = "README.md"
-requires-python = ">=3.12"
+requires-python = ">=3.10"
 dependencies = [
     "bs4>=0.0.2",
     "gradio[cli]>=5.33.1",
@@ -13,5 +13,6 @@ dependencies = [
     "pillow>=11.2.1",
     "torch[cuda]>=2.7.1",
     "transformers>=4.52.4",
+    "ultralytics>=8.0.0",
 ]
 

yolov8x-world.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b99398e46cffbf2b9a7e668512fa295f0d710d173ae0a815ec706ced5d1099b
+size 147961954