Update app.py

app.py

@@ -1,3 +1,8 @@
+# Install cloudpickle first
+import subprocess
+subprocess.run(["pip", "install", "cloudpickle", "-q"], check=True)
+import cloudpickle
+
 # Patch Pipeline
 import fasttransform
 import fastcore.transform as _fct
@@ -30,22 +35,38 @@ def load_learner_fixed(fname):
 
 # Load model
 model = load_learner_fixed("clothing_classifier.pkl")
-labels = model.dls.vocab
+all_labels = model.dls.vocab
 
-# Prediction function — MultiCategoryBlock needs sigmoid threshold
+# Prediction function
 def predict(img):
     img = PILImage.create(img)
     pred, pred_idx, probs = model.predict(img)
-    # Convert tensor to plain float dict
-    result = {}
-    for i, label in enumerate(labels):
-        result[str(label)] = float(probs[i])
+
+    age_labels      = ['Kids', 'Adults']
+    clothing_labels = [l for l in all_labels if l not in age_labels]
+
+    # Get best clothing label from all probs (not just threshold-filtered pred)
+    clothing_probs = {l: float(probs[list(all_labels).index(l)]) for l in clothing_labels}
+    best_clothing  = max(clothing_probs, key=clothing_probs.get)
+
+    # Get age
+    kids_prob   = float(probs[list(all_labels).index('Kids')])
+    adults_prob = float(probs[list(all_labels).index('Adults')])
+    age = 'Kids' if kids_prob > adults_prob else 'Adults'
+
+    # Return top clothing labels only (no Adults/Kids in output)
+    top_clothing = dict(sorted(clothing_probs.items(), key=lambda x: x[1], reverse=True)[:5])
+    
+    # Add age info as first key
+    result = {f"✅ {best_clothing} for {age}": 1.0}
+    result.update(top_clothing)
+    
     return result
 
 demo = gr.Interface(
     fn=predict,
     inputs=gr.Image(type="pil"),
-    outputs=gr.Label(num_top_classes=5),
+    outputs=gr.Label(num_top_classes=6),
     title="👗 Clothes Classifier",
     description="Upload a clothing image and the model will classify it!"
 )