Update app.py

app.py

@@ -5,12 +5,13 @@ from huggingface_hub import hf_hub_download
 import tensorflow as tf
 from tensorflow.keras.applications.efficientnet import preprocess_input
 
-# ── Config ────────────────────────────────────────────────
-MODEL_REPO       = "Unearthed/artifact-classifier"
-MODEL_FILE       = "artifact_classifier.keras"
-LABELS_FILE      = "labels.txt"
-IMG_SIZE         = (224, 224)
+# ── Config ───────────────────────────────────────────────
+MODEL_REPO           = "Unearthed/artifact-classifier"
+MODEL_FILE           = "artifact_classifier.keras"
+LABELS_FILE          = "labels.txt"
+IMG_SIZE             = (224, 224)
 CONFIDENCE_THRESHOLD = 0.60
+TEXTURE_THRESHOLD    = 0.15
 # ─────────────────────────────────────────────────────────
 
 print("Loading model...")
@@ -25,26 +26,41 @@ print("Classes:", class_names)
 print("Model ready!")
 
 
+def texture_score(img_array):
+    gray = np.mean(img_array, axis=2)
+    patch_size = 8
+    h, w = gray.shape
+    variances = []
+    for i in range(0, h - patch_size, patch_size):
+        for j in range(0, w - patch_size, patch_size):
+            patch = gray[i:i+patch_size, j:j+patch_size]
+            variances.append(np.var(patch))
+    return min(np.mean(variances) / 300.0, 1.0)
+
+
 def predict(image):
     if image is None:
         return "Please upload an image.", {}
 
     img = image.convert("RGB").resize(IMG_SIZE)
-    x = np.array(img, dtype=np.float32)
-    x = preprocess_input(x)
-    x = np.expand_dims(x, 0)
+    img_array = np.array(img, dtype=np.float32)
 
+    tex = texture_score(img_array)
+    if tex < TEXTURE_THRESHOLD:
+        return f"Not an artifact — no stone texture detected (texture score: {tex:.2f})", {"notartifact": 1.0}
+
+    x = preprocess_input(img_array.copy())
+    x = np.expand_dims(x, 0)
     preds = model.predict(x, verbose=0)[0]
 
     top_idx   = int(np.argmax(preds))
     top_score = float(preds[top_idx])
-
-    scores = {class_names[i]: float(preds[i]) for i in range(len(preds))}
+    scores    = {class_names[i]: float(preds[i]) for i in range(len(preds))}
 
     if top_score < CONFIDENCE_THRESHOLD:
-        label = f"Not an artifact  ({top_score*100:.1f}% confidence — below {CONFIDENCE_THRESHOLD*100:.0f}% threshold)"
+        label = f"Not an artifact ({top_score*100:.1f}% confidence — below {CONFIDENCE_THRESHOLD*100:.0f}% threshold)"
     else:
-        label = f"{class_names[top_idx]}  ({top_score*100:.1f}% confidence)"
+        label = f"{class_names[top_idx]} ({top_score*100:.1f}% confidence)"
 
     return label, scores