final tuning for best results

.gitattributes

@@ -19,7 +19,6 @@
 *.pb filter=lfs diff=lfs merge=lfs -text
 *.pickle filter=lfs diff=lfs merge=lfs -text
 *.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
 *.pth filter=lfs diff=lfs merge=lfs -text
 *.rar filter=lfs diff=lfs merge=lfs -text
 *.safetensors filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -25,7 +25,7 @@ logs/
 *.swo
 
 # Ignore dataset and large model files
-./app/models/
+app/models/
 
 # Ignore temp files generated during inference
 temp_*.png

app/main.py

@@ -18,7 +18,7 @@ logger = logging.getLogger(__name__)
 # Enable CORS
 app.add_middleware(
     CORSMiddleware,
-  allow_origins=["https://ai-powered-skin-care-ecommerce.vercel.app"],
+  allow_origins=["https://ai-powered-skin-care-ecommerce.vercel.app","http://localhost:5173"],
     allow_credentials=True,
     allow_methods=["*"],
     allow_headers=["*"],

app/services/acne_service.py

@@ -13,7 +13,7 @@ MODEL_PATH = "app/models/acne_model.pt"
 model = AutoDetectionModel.from_pretrained(
      model_type='yolov8',
     model_path=MODEL_PATH,
-    confidence_threshold=0.3,
+    confidence_threshold=0.25,
     device="cpu"
 )
 
@@ -26,11 +26,11 @@ def predict_acne(image: Image.Image):
 
 
     # Dynamically calculate slice size (approx. 1/3rd of the image dimension)
-    slice_height = 256
-    slice_width = 256
+    slice_height = 320
+    slice_width = 320
     # Dynamically calculate overlap (e.g., 20% of slice size as ratio to image size)
-    overlap_height_ratio = 0.2
-    overlap_width_ratio = 0.2
+    overlap_height_ratio = 0.3
+    overlap_width_ratio = 0.3
 
     logging.info(f"[SAHI] Using slice size ({slice_width}x{slice_height}) with overlap ({overlap_width_ratio:.2f}, {overlap_height_ratio:.2f})")
 

app/services/puffy_eyes_service.py

@@ -10,11 +10,10 @@ model = YOLO(MODEL_PATH)
 
 # Set confidence threshold
 CONFIDENCE_THRESHOLD = 0.25
-MAX_SHOW = 2  # only show up to 2 detections
+MAX_SHOW = 2  # Only allow 2 detections
 
 def predict_puffy_eyes(image: Image):
-    """Run YOLO model on the input image, draw up to MAX_SHOW highest-confidence boxes,
-       and return detections + processed image plus max_confidence if >2 initial boxes."""
+    """Run YOLO model, only return if exactly 2 detections."""
     
     # Run YOLO detection
     results = model(image, conf=CONFIDENCE_THRESHOLD)
@@ -27,30 +26,31 @@ def predict_puffy_eyes(image: Image):
                 x1, y1, x2, y2 = [float(c) for c in box.xyxy[0]]
                 all_dets.append({"confidence": conf, "bbox": [x1, y1, x2, y2]})
 
-    # Sort by confidence descending
+    # Only continue if exactly 2 detections
+    if len(all_dets) != 2:
+     logging.info(f"[Puffy Eyes] Skipping: found {len(all_dets)} detections (need exactly 2)")
+     return {
+        "detections": [],
+        "labeled_image": None,
+        "max_confidence": None,
+        "message": f"Expected 2 puffy eye detections, but found {len(all_dets)}"
+        }
+
+    # Sort by confidence
     all_dets.sort(key=lambda d: d["confidence"], reverse=True)
 
-    # Keep only top MAX_SHOW
-    keep = all_dets[:MAX_SHOW]
-    extra = all_dets[MAX_SHOW:]
-    if extra:
-        max_conf = extra[0]["confidence"]
-    else:
-        max_conf = None
-
-    # Draw only kept boxes
+    # Draw the two boxes
     draw = ImageDraw.Draw(image)
-    for det in keep:
+    for det in all_dets:
         x1, y1, x2, y2 = det["bbox"]
         draw.rectangle([x1, y1, x2, y2], outline="red", width=3)
 
-    
     # Encode annotated image
     buf = io.BytesIO()
     image.save(buf, format="JPEG")
     encoded_image = base64.b64encode(buf.getvalue()).decode("utf-8")
 
-    out = {"detections": keep, "labeled_image": encoded_image}
-    if max_conf is not None:
-        out["max_confidence"] = max_conf
-    return out
+    return {
+        "detections": all_dets,
+        "labeled_image": encoded_image
+    }

app/services/skin_analysis_service.py

@@ -16,7 +16,7 @@ from app.services.acne_severity_service import predict_acne_severity
 from app.services.skin_type_service import predict_skin_type
 
 # Define the size to which images will be resized for model input
-MODEL_INPUT_SIZE = (512, 512)
+MODEL_INPUT_SIZE = (640, 640)
 
 def skin_analysis(image: Image.Image) -> Dict[str, Any]:
     try: