Update app.py

app.py

@@ -2,71 +2,94 @@ import pandas as pd
 import torch
 from transformers import CLIPProcessor, CLIPModel
 from sklearn.metrics.pairwise import cosine_similarity
-from PIL import Image
+from PIL import Image, UnidentifiedImageError
 import gradio as gr
 from pathlib import Path
+import os
 
-# Carica modello e processor CLIP
+# Imposta device
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+# Carica modello CLIP
 def load_clip_model(device):
-    model = CLIPModel.from_pretrained("laion/CLIP-ViT-L-14-laion2B-s32B-b82K").to(device)
-    processor = CLIPProcessor.from_pretrained("laion/CLIP-ViT-L-14-laion2B-s32B-b82K")
-    return model, processor
+    try:
+        model = CLIPModel.from_pretrained("laion/CLIP-ViT-L-14-laion2B-s32B-b82K").to(device)
+        processor = CLIPProcessor.from_pretrained("laion/CLIP-ViT-L-14-laion2B-s32B-b82K")
+        return model, processor
+    except Exception as e:
+        print("Errore nel caricamento del modello CLIP:", e)
+        raise e
 
-# Carica embedding da CSV
+# Carica embeddings dal CSV
 def load_embeddings(embedding_file):
-    df = pd.read_csv(embedding_file)
-    embeddings = df.iloc[:, 1:].values  # Esclude la colonna 'filename'
-    image_paths = df['filename'].tolist()
-    return embeddings, image_paths
+    try:
+        df = pd.read_csv(embedding_file)
+        assert 'filename' in df.columns, "La colonna 'filename' è obbligatoria nel CSV"
+        embeddings = df.drop(columns=['filename']).values
+        image_paths = df['filename'].tolist()
+        return embeddings, image_paths
+    except Exception as e:
+        print("Errore nel caricamento degli embeddings:", e)
+        raise e
 
 # Trova immagini simili
 def query_images(text, model, processor, image_embeddings, image_paths, device):
-    text_inputs = processor(text=[text], return_tensors="pt", padding=True).to(device)
-    with torch.no_grad():
-        text_embedding = model.get_text_features(**text_inputs).cpu().numpy().flatten()
+    try:
+        text_inputs = processor(text=[text], return_tensors="pt", padding=True).to(device)
+        with torch.no_grad():
+            text_embedding = model.get_text_features(**text_inputs).cpu().numpy().flatten()
 
-    similarities = cosine_similarity([text_embedding], image_embeddings)[0]
-    top_indices = similarities.argsort()[-3:][::-1]
-    return [(Path("img") / image_paths[i], similarities[i]) for i in top_indices]
+        similarities = cosine_similarity([text_embedding], image_embeddings)[0]
+        top_indices = similarities.argsort()[-3:][::-1]
+        return [(Path("img") / image_paths[i], similarities[i]) for i in top_indices]
+    except Exception as e:
+        print("Errore nel calcolo delle similarità:", e)
+        return []
 
-# Funzione per Gradio
+# Funzione di predizione
 def predict(query_text):
-    similar_images = query_images(query_text, model, processor, embeddings, image_paths, device)
-    image_outputs = []
-    scores = []
+    try:
+        similar_images = query_images(query_text, model, processor, embeddings, image_paths, device)
+        image_outputs = []
+        scores = []
+
+        for img_path, score in similar_images:
+            try:
+                img = Image.open(img_path).convert("RGB")
+                image_outputs.append(img)
+                scores.append(score)
+            except (FileNotFoundError, UnidentifiedImageError) as e:
+                print(f"Errore nell'apertura immagine {img_path}: {e}")
+                continue
 
-    for img_path, score in similar_images:
-        try:
-            img = Image.open(img_path).convert("RGB")
-            image_outputs.append(img)
-            scores.append(score)
-        except Exception as e:
-            print(f"Errore nell'apertura immagine {img_path}: {e}")
-            continue
+        if not image_outputs:
+            # Nessuna immagine caricabile
+            return [], pd.DataFrame([["Nessuna immagine trovata"]], columns=["Errore"])
 
-    df_scores = pd.DataFrame(scores, columns=["Similarity Score"])
-    return image_outputs, df_scores
+        df_scores = pd.DataFrame(scores, columns=["Similarity Score"])
+        return image_outputs, df_scores
+    except Exception as e:
+        print("Errore durante la predizione:", e)
+        return [], pd.DataFrame([["Errore interno"]], columns=["Errore"])
 
-# Avvio Gradio
+# Esecuzione
 if __name__ == "__main__":
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    model, processor = load_clip_model(device)
+    try:
+        model, processor = load_clip_model(device)
+        embeddings, image_paths = load_embeddings("embeddings.csv")
 
-    # Carica embedding
-    embedding_file = "embeddings.csv"
-    embeddings, image_paths = load_embeddings(embedding_file)
+        interface = gr.Interface(
+            fn=predict,
+            inputs=gr.Textbox(label="Inserisci il testo"),
+            outputs=[
+                gr.Gallery(label="Immagini simili"),
+                gr.Dataframe(label="Punteggi di similarità")
+            ],
+            title="Ricerca immagini simili con CLIP",
+            description="Inserisci un testo per trovare le immagini più affini nel database."
+        )
 
-    # Interfaccia Gradio
-    interface = gr.Interface(
-        fn=predict,
-        inputs=gr.Textbox(label="Enter your text"),
-        outputs=[
-            gr.Gallery(label="Top 3 Similar Images"),
-            gr.Dataframe(label="Similarity Scores")
-        ],
-        title="CLIP Image Finder",
-        description="Enter a textual description to find the most similar images using CLIP."
-    )
+        interface.launch(share=True)
 
-    # Per Hugging Face è obbligatorio `share=True`
-    interface.launch(share=True)
+    except Exception as e:
+        print("Errore durante l'inizializzazione dell'app:", e)