Update app.py

app.py

@@ -1,134 +1,114 @@
+
 import gradio as gr
 import torch
 import pandas as pd
 import numpy as np
-from transformers import CLIPModel, CLIPProcessor
 from PIL import Image
-import io
+from transformers import CLIPProcessor, CLIPModel
+from datasets import load_dataset
+from torch.nn import functional as F
 
-# --- CONFIGURATION ---
+# --- 1. SETUP & CONFIG ---
 MODEL_ID = "openai/clip-vit-base-patch32"
 DATA_FILE = "food_embeddings_clip.parquet"
 
-# 🎥 PASTE YOUR YOUTUBE VIDEO ID HERE
-# (e.g. if link is https://www.youtube.com/watch?v=dQw4w9WgXcQ, the ID is dQw4w9WgXcQ)
-YOUTUBE_ID = "IXeIxYHi0Es" 
-
-print(f"⏳ Loading {MODEL_ID} and Data...")
-
-# 1. Load Model
+print("⏳ Starting App... Loading Model...")
+# Load Model (CPU is fine for inference on single images)
 model = CLIPModel.from_pretrained(MODEL_ID)
 processor = CLIPProcessor.from_pretrained(MODEL_ID)
 
-# 2. Load Data
-try:
-    df = pd.read_parquet(DATA_FILE)
-    # Prepare Vectors
-    all_vectors = np.stack(df['embedding'].to_numpy())
-    db_features = torch.tensor(all_vectors)
-except FileNotFoundError:
-    raise RuntimeError(f"❌ ERROR: Could not find {DATA_FILE}. Did you upload it?")
-
-print("✅ System Ready!")
-
-# --- SEARCH LOGIC ---
-def search(text_query, image_query):
-    if not text_query and not image_query:
-        return []
-
-    # A. Determine Input
-    if image_query:
-        inputs = processor(images=image_query, return_tensors="pt", padding=True)
-        get_feat = model.get_image_features
-    else:
-        inputs = processor(text=[text_query], return_tensors="pt", padding=True)
-        get_feat = model.get_text_features
-
-    # B. Inference & Search
-    with torch.no_grad():
-        query_vec = get_feat(**inputs)
-        top_scores, top_indices = torch.topk(scores, k=5)
-
-    # C. Format Results
+# --- 2. LOAD DATA (Must match Colab logic EXACTLY) ---
+print("⏳ Loading Dataset (this takes a moment)...")
+# We load the same 5000 images using the same seed so indices match the parquet file
+dataset = load_dataset("ethz/food101", split="train").shuffle(seed=42).select(range(5000))
+
+# --- 3. LOAD EMBEDDINGS ---
+print("⏳ Loading Pre-computed Embeddings...")
+df = pd.read_parquet(DATA_FILE)
+# Convert the list of numbers in the parquet back to a Torch Tensor
+db_features = torch.tensor(np.stack(df['embedding'].to_numpy()))
+# Normalize once for speed
+db_features = F.normalize(db_features, p=2, dim=1)
+
+print("✅ App Ready!")
+
+# --- 4. CORE SEARCH LOGIC ---
+def find_best_matches(query_features, top_k=3):
+    # Normalize query
+    query_features = F.normalize(query_features, p=2, dim=1)
+    
+    # Calculate Similarity (Dot Product)
+    # Query (1x512) * DB (5000x512) = Scores (1x5000)
+    similarity = torch.mm(query_features, db_features.T)
+    
+    # Get Top K
+    scores, indices = torch.topk(similarity, k=top_k)
+    
     results = []
-    for idx, score in zip(top_indices[0], top_scores[0]):
-        row = df.iloc[idx.item()]
+    for idx, score in zip(indices[0], scores[0]):
+        idx = idx.item()
         
-        # Load Image
-        img_data = row['image']
-        if isinstance(img_data, dict) and 'bytes' in img_data:
-            img = Image.open(io.BytesIO(img_data['bytes']))
-        else:
-            img = img_data
-
-        results.append((img, f"{row['label_name']} ({score.item():.2f})"))
+        # Grab image and info from the loaded dataset
+        img = dataset[idx]['image']
+        label = df.iloc[idx]['label_name'] # Get label from our dataframe
+        
+        # Format output
+        results.append((img, f"{label} ({score:.2f})"))
     return results
 
-# --- APP INTERFACE (The Original Design) ---
-# We use a 'Soft' theme for a professional look
-with gr.Blocks(theme=gr.themes.Soft(), title="AI Food Search") as demo:
+# --- 5. GRADIO FUNCTIONS ---
+def search_by_image(input_image):
+    if input_image is None: return []
     
-    # 1. Header Section
-    gr.Markdown(
-        """
-        # 🍔 AI Food Search Engine
-        ### Powered by OpenAI CLIP & Hugging Face
-        Search through 5,000 food images using natural language or reference images.
-        """
-    )
-    
-    # 2. YouTube Demo Section (Embedded Player)
-    if YOUTUBE_ID and YOUTUBE_ID != "YOUR_YOUTUBE_ID_HERE":
-        gr.HTML(
-            f"""
-            <div style="display: flex; justify-content: center; margin-bottom: 20px;">
-                <iframe width="560" height="315" 
-                src="https://www.youtube.com/embed/{YOUTUBE_ID}" 
-                title="YouTube video player" frameborder="0" 
-                allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" 
-                allowfullscreen></iframe>
-            </div>
-            """
-        )
-    else:
-        gr.Info("ℹ️ Add your YouTube ID in the code to display the video here.")
-
-    # 3. Main Search Interface
-    with gr.Row():
-        # Left Column: Inputs
-        with gr.Column(scale=1):
-            gr.Markdown("### 🔍 Your Query")
-            txt_input = gr.Textbox(
-                label="Search by Text", 
-                placeholder="e.g. 'spicy tacos with lime'",
-                show_label=True
-            )
-            gr.Markdown("**OR**")
-            img_input = gr.Image(
-                type="pil", 
-                label="Search by Image",
-                height=300
-            )
-            
-            search_btn = gr.Button("🚀 Find Food", variant="primary", size="lg")
-
-        # Right Column: Results
-        with gr.Column(scale=2):
-            gr.Markdown("### 🍕 Top Matches")
-            gallery = gr.Gallery(
-                label="Results", 
-                columns=3, 
-                height="auto",
-                object_fit="cover"
-            )
-
-    # 4. Footer / Credits
-    gr.Markdown("---")
-    gr.Markdown(f"*Model: {MODEL_ID} | Dataset: Food-101 (Subset)*")
+    inputs = processor(images=input_image, return_tensors="pt")
+    with torch.no_grad():
+        features = model.get_image_features(**inputs)
+        
+    return find_best_matches(features)
 
-    # Event Listeners (Enter Key + Button Click)
-    txt_input.submit(search, inputs=[txt_input, img_input], outputs=gallery)
-    search_btn.click(search, inputs=[txt_input, img_input], outputs=gallery)
+def search_by_text(input_text):
+    if not input_text: return []
+    
+    inputs = processor(text=[input_text], return_tensors="pt", padding=True)
+    with torch.no_grad():
+        features = model.get_text_features(**inputs)
+        
+    return find_best_matches(features)
 
-# Launch
-demo.launch(server_name="0.0.0.0", server_port=7860)
+# --- 6. BUILD UI ---
+with gr.Blocks(title="Food Matcher AI") as demo:
+    gr.Markdown("# 🍔 Visual Dish Matcher")
+    gr.Markdown("Upload a photo of food (or describe it) to find similar dishes in our database.")
+    
+    # --- VIDEO SECTION ---
+    # Using Accordion so it doesn't clutter the UI. Open=False means it starts closed.
+    with gr.Accordion("📺 Watch Project Demo", open=False):
+        gr.HTML("""
+            <div style="display: flex; justify-content: center;">
+                <iframe width="560" height="315"
+                    src="https://www.youtube.com/embed/IXeIxYHi0Es"
+                    title="YouTube video player"
+                    frameborder="0"
+                    allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
+                    allowfullscreen>
+                </iframe>
+            </div>
+        """)
+    # ----------------------------
+    
+    with gr.Tab("Image Search"):
+        with gr.Row():
+            img_input = gr.Image(type="pil", label="Upload Food Image")
+            img_gallery = gr.Gallery(label="Top Matches")
+        btn_img = gr.Button("Find Similar Dishes")
+        btn_img.click(search_by_image, inputs=img_input, outputs=img_gallery)
+
+    with gr.Tab("Text Search"):
+        with gr.Row():
+            txt_input = gr.Textbox(label="Describe the food (e.g., 'Spicy Tacos')")
+            txt_gallery = gr.Gallery(label="Top Matches")
+        btn_txt = gr.Button("Search by Description")
+        btn_txt.click(search_by_text, inputs=txt_input, outputs=txt_gallery)
+
+# Launch (Disable SSR for stability)
+demo.launch(ssr_mode=False)