Update app.py

app.py

@@ -1,79 +1,103 @@
 import gradio as gr
 import torch
 import numpy as np
-from collections import defaultdict
+from PIL import Image
 from transformers import AutoImageProcessor, AutoModel
-from sklearn.metrics.pairwise import cosine_similarity
-
-# Load model pretrained (visual embedding)
-processor = AutoImageProcessor.from_pretrained("facebook/dino-vits16")
-model = AutoModel.from_pretrained("facebook/dino-vits16")
+import os
+
+# =========================
+# Load model pretrained
+# =========================
+processor = AutoImageProcessor.from_pretrained(
+    "google/vit-base-patch16-224"
+)
+model = AutoModel.from_pretrained(
+    "google/vit-base-patch16-224"
+)
 model.eval()
 
-# Database: label -> list of features
-feature_db = defaultdict(list)
-
-def extract_feature(image):
+# =========================
+# Ekstraksi fitur
+# =========================
+def get_embedding(image):
     inputs = processor(images=image, return_tensors="pt")
     with torch.no_grad():
         outputs = model(**inputs)
-    return outputs.last_hidden_state.mean(dim=1).numpy()
-
-def tambah_sampling(image, label):
-    if image is None or label.strip() == "":
-        return "❌ Gambar dan label wajib diisi"
-
-    feat = extract_feature(image)
-    feature_db[label].append(feat)
-
-    return f"✅ Sampling ditambahkan untuk label '{label}' (total: {len(feature_db[label])})"
-
-def tebak_gambar(image):
-    if image is None:
-        return "❌ Upload gambar terlebih dahulu"
-
-    if len(feature_db) == 0:
-        return "❌ Belum ada data sampling"
-
-    test_feat = extract_feature(image)
-
-    label_scores = {}
-    for label, feats in feature_db.items():
-        sims = cosine_similarity(test_feat, np.vstack(feats))
-        label_scores[label] = sims.mean()  # rata-rata kemiripan
-
-    # Ambil label dengan skor tertinggi
-    pred_label = max(label_scores, key=label_scores.get)
-    skor = label_scores[pred_label]
-
-    hasil = "📌 HASIL PREDIKSI (BERDASARKAN KEMIRIPAN)\n\n"
-    for lbl, sc in sorted(label_scores.items(), key=lambda x: x[1], reverse=True):
-        hasil += f"- {lbl}: {sc:.2f}\n"
-
-    hasil += f"\n✅ Gambar PALING MIRIP dengan: **{pred_label}**"
-
+    embedding = outputs.last_hidden_state.mean(dim=1)
+    return embedding.squeeze().numpy()
+
+# =========================
+# Cosine similarity
+# =========================
+def cosine_similarity(a, b):
+    return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))
+
+# =========================
+# Fungsi utama
+# =========================
+def unsupervised_image_similarity(sample_path, test_path):
+    if sample_path is None or test_path is None:
+        return "❌ Silakan upload gambar sampling dan gambar uji."
+
+    # Ambil nama file sampling
+    sample_name = os.path.basename(sample_path)
+
+    # Buka gambar
+    img_sample = Image.open(sample_path).convert("RGB")
+    img_test = Image.open(test_path).convert("RGB")
+
+    # Ekstraksi embedding
+    emb1 = get_embedding(img_sample)
+    emb2 = get_embedding(img_test)
+
+    similarity = cosine_similarity(emb1, emb2)
+
+    # Threshold
+    threshold = 0.75
+
+    if similarity >= threshold:
+        status = (
+            f"✅ GAMBAR UJI MEMILIKI KEMIRIPAN\n"
+            f"dengan SAMPLING: **{sample_name}**"
+        )
+    else:
+        status = (
+            "❌ GAMBAR UJI TIDAK MEMILIKI KEMIRIPAN\n"
+            "dengan sampling yang diinput"
+        )
+
+    hasil = f"""
+HASIL UNSUPERVISED LEARNING
+
+Nama sampling  : {sample_name}
+Nilai kemiripan: {similarity:.2f}
+
+Keputusan:
+{status}
+
+Catatan:
+- Tidak menggunakan dataset
+- Tidak menggunakan label
+- Tidak melakukan training ulang
+- Pendekatan unsupervised berbasis similarity
+"""
     return hasil
 
-with gr.Blocks() as app:
-    gr.Markdown("## 🧠 Tebak Gambar Supervised (Similarity-Based)")
-    gr.Markdown(
-        "Upload gambar sampling dengan label, lalu upload gambar baru. "
-        "AI menentukan label berdasarkan kemiripan visual, bukan gambar yang sama persis."
+# =========================
+# UI Gradio
+# =========================
+app = gr.Interface(
+    fn=unsupervised_image_similarity,
+    inputs=[
+        gr.Image(type="filepath", label="Gambar Sampling"),
+        gr.Image(type="filepath", label="Gambar Uji")
+    ],
+    outputs=gr.Textbox(label="Hasil Analisis"),
+    title="🧠 Unsupervised Image Similarity dengan Nama Sampling",
+    description=(
+        "Aplikasi unsupervised learning untuk mengecek apakah gambar uji "
+        "memiliki kemiripan dengan gambar sampling tertentu."
     )
-
-    with gr.Tab("1️⃣ Tambah Data Sampling"):
-        img1 = gr.Image(type="pil", label="Gambar Sampling")
-        lbl = gr.Textbox(label="Label")
-        out1 = gr.Textbox(label="Status")
-        btn1 = gr.Button("Tambah")
-
-        btn1.click(tambah_sampling, [img1, lbl], out1)
-
-    with gr.Tab("2️⃣ Tebak Gambar"):
-        img2 = gr.Image(type="pil", label="Gambar Uji")
-        out2 = gr.Textbox(label="Hasil")
-        btn2 = gr.Button("Tebak")
-
-        btn2.click(tebak_gambar, img2, out2)
+)
 
 app.launch()