Adjust constants

Co-authored-by: aider (gemini/gemini-2.5-pro-preview-05-06) <aider@aider.chat>

.gitignore

@@ -150,3 +150,4 @@ out.*
 *.csv
 *.txt
 *.json
+.aider*

model.py

@@ -5,7 +5,7 @@ from PIL import Image, ImageDraw
 # For the custom crop transform. User can adjust these.
 TRIANGLE_CROP_WIDTH = 556
 TRIANGLE_CROP_HEIGHT = 1184
-RESIZE_DIM = 64 # Resize cropped image to this dimension (square)
+RESIZE_DIM = 256 # Resize cropped image to this dimension (square)
 
 
 # Custom transform to crop a triangle from the lower right corner

move.py

@@ -10,8 +10,8 @@ def main():
     them into data/open/ or data/closed/ based on the 'choice' field.
     """
     data_dir = 'data'
-    open_dir = os.path.join(data_dir, 'open')
+    open_dir = os.path.join(data_dir, 'labelled/open')
-    closed_dir = os.path.join(data_dir, 'closed')
+    closed_dir = os.path.join(data_dir, 'labelled/closed')
     
     # This is an assumption based on the path found in the 'd' parameter of the image URL.
     # e.g., ...?d=.../data/hourly_photos/filename.jpg

server.py

@@ -0,0 +1,137 @@
+import asyncio
+import aiohttp
+from aiohttp import web
+import logging
+import os
+import io
+
+import torch
+import torch.nn.functional as F
+from torchvision import transforms
+from PIL import Image
+
+from model import (CropLowerRightTriangle, GarageDoorCNN, TRIANGLE_CROP_WIDTH,
+                   TRIANGLE_CROP_HEIGHT, RESIZE_DIM)
+
+# --- Configuration ---
+logging.basicConfig(level=logging.DEBUG, format='%(asctime)s - %(levelname)s - %(message)s')
+
+BLUEIRIS_KEY = os.getenv('BLUEIRIS_KEY')
+if not BLUEIRIS_KEY:
+    raise ValueError("BLUEIRIS_KEY environment variable not set.")
+
+CAMERA_URL = "http://cameras.dns.t0.vc/image/SE-S?&w=9999&decode=1"
+MODEL_PATH = 'garage_door_cnn.pth'
+CLASS_NAMES = ['closed', 'open'] # From training, sorted alphabetically
+POLL_INTERVAL_SECONDS = 10
+REQUEST_TIMEOUT_SECONDS = 5
+
+# --- Model Inference ---
+def get_prediction(model, image_bytes, device):
+    """Run model inference on the provided image bytes."""
+    try:
+        image = Image.open(io.BytesIO(image_bytes)).convert('RGB')
+    except Exception as e:
+        logging.error(f"Failed to open image from bytes: {e}")
+        return None
+
+    # Define the same transforms as used in validation
+    transform = transforms.Compose([
+        CropLowerRightTriangle(triangle_width=TRIANGLE_CROP_WIDTH, triangle_height=TRIANGLE_CROP_HEIGHT),
+        transforms.Resize((RESIZE_DIM, RESIZE_DIM)),
+        transforms.ToTensor(),
+        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+    ])
+
+    input_tensor = transform(image).unsqueeze(0).to(device)
+
+    with torch.no_grad():
+        output = model(input_tensor)
+        probabilities = F.softmax(output, dim=1)
+        confidence, pred_idx = torch.max(probabilities, 1)
+        return CLASS_NAMES[pred_idx.item()], confidence.item()
+
+# --- Background Task ---
+async def monitor_garage_door(app):
+    """Periodically fetches an image and logs the garage door status."""
+    logging.info("Starting garage door monitoring task.")
+    session = app['client_session']
+    model = app['model']
+    device = app['device']
+    headers = {'Authorization': 'Basic ' + BLUEIRIS_KEY}
+
+    while True:
+        try:
+            await asyncio.sleep(POLL_INTERVAL_SECONDS)
+            logging.debug("Fetching new image from camera...")
+            async with session.get(CAMERA_URL, headers=headers, timeout=REQUEST_TIMEOUT_SECONDS) as response:
+                if response.status == 200:
+                    image_bytes = await response.read()
+                    result = get_prediction(model, image_bytes, device)
+                    if result:
+                        prediction, confidence = result
+                        logging.debug(f"Garage door status: {prediction} (confidence: {confidence:.4f})")
+                else:
+                    logging.error(f"Failed to fetch image. Status: {response.status}, Reason: {response.reason}")
+
+        except asyncio.TimeoutError:
+            logging.warning("Request to camera timed out.")
+        except aiohttp.ClientError as e:
+            logging.error(f"Client error during image fetch: {e}")
+        except asyncio.CancelledError:
+            logging.info("Monitoring task cancelled.")
+            break
+        except Exception as e:
+            logging.error(f"An unexpected error occurred in the monitoring task: {e}", exc_info=True)
+            # Add a small delay before retrying on unexpected errors
+            await asyncio.sleep(5)
+
+
+# --- Web Server ---
+async def handle_root(request):
+    """Handler for the root GET request."""
+    return web.Response(text="hello world")
+
+async def on_startup(app):
+    """Actions to perform on application startup."""
+    # Set up device
+    app['device'] = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    logging.info(f"Using device: {app['device']}")
+
+    # Load model
+    logging.info(f"Loading model from {MODEL_PATH}...")
+    model = GarageDoorCNN(resize_dim=RESIZE_DIM).to(app['device'])
+    model.load_state_dict(torch.load(MODEL_PATH, map_location=app['device']))
+    model.eval()
+    app['model'] = model
+    logging.info("Model loaded successfully.")
+
+    # Create client session
+    app['client_session'] = aiohttp.ClientSession()
+
+    # Start background task
+    app['monitor_task'] = asyncio.create_task(monitor_garage_door(app))
+
+async def on_cleanup(app):
+    """Actions to perform on application cleanup."""
+    logging.info("Cleaning up...")
+    app['monitor_task'].cancel()
+    try:
+        await app['monitor_task']
+    except asyncio.CancelledError:
+        pass
+    await app['client_session'].close()
+    logging.info("Cleanup complete.")
+
+def main():
+    app = web.Application()
+    app.router.add_get('/', handle_root)
+    app.on_startup.append(on_startup)
+    app.on_cleanup.append(on_cleanup)
+    web.run_app(app, port=8081)
+
+if __name__ == '__main__':
+    if not os.path.exists(MODEL_PATH):
+        logging.error(f"Model file '{MODEL_PATH}' not found. Please run train.py first.")
+    else:
+        main()

sort.py

@@ -14,7 +14,7 @@ def sort_images():
     MODEL_PATH = 'garage_door_cnn.pth'
     SOURCE_DIR = 'data/hourly_photos/'
     DEST_DIR = 'data/sorted/open/'
-    CONFIDENCE_THRESHOLD = 0.90 # Only copy if confidence is over this value
+    CONFIDENCE_THRESHOLD = 0.80 # Only copy if confidence is over this value
     
 
     # The classes are sorted alphabetically by ImageFolder: ['closed', 'open']