app.py

import tensorflow as tf
import gradio as gr
import numpy as np
import os
import warnings
from PIL import Image

warnings.filterwarnings("ignore")

# ============================================================
# 1. LOAD MODEL
# ============================================================
print("=" * 60)
print("LOADING MODEL")
print("=" * 60)

MODEL_PATH = "model.keras"

try:
    best_model = tf.keras.models.load_model(
        MODEL_PATH,
        custom_objects={"quantization_config": None},
        compile=False
    )
    print("✅ Model berhasil dimuat")
except Exception as e:
    print("❌ Pemuatan model gagal, membuat model dummy:", e)

    from tensorflow.keras import layers, Model
    inputs = layers.Input(shape=(224, 224, 3))
    x = layers.GlobalAveragePooling2D()(inputs)
    dr_output = layers.Dense(5, name="dr_head")(x)    # logits
    dme_output = layers.Dense(3, name="dme_head")(x)  # logits
    best_model = Model(inputs, {"dr_head": dr_output, "dme_head": dme_output})
    best_model.compile(optimizer="adam", loss="categorical_crossentropy")

best_model.summary()

# ============================================================
# 2. CONFIG
# ============================================================
IMG_SIZE = 224

# ============================================================
# 3. PREPROCESSING
# ============================================================
def preprocess_image(img_path):
    img = tf.io.read_file(img_path)
    img = tf.image.decode_jpeg(img, channels=3)
    img = tf.image.resize(img, (IMG_SIZE, IMG_SIZE))
    img = tf.cast(img, tf.float32) / 255.0
    return tf.expand_dims(img, 0)

# ============================================================
# 4. CLASS MAPPING
# ============================================================
DR_CLASSES = ["No DR", "Mild", "Moderate", "Severe", "Proliferative DR"]
DME_CLASSES = ["No DME", "Low Risk", "High Risk"]

# ============================================================
# 5. SOFTMAX SAFETY
# ============================================================
def ensure_probability(x):
    x = np.asarray(x, dtype=np.float32)
    if x.min() < 0 or x.max() > 1.0 or abs(x.sum() - 1.0) > 1e-3:
        x = tf.nn.softmax(x).numpy()
    return x

# ============================================================
# 6. PREDICTION → TABLE ONLY
# ============================================================
def predict_and_format_text(image_path):
    img_tensor = preprocess_image(image_path)
    preds = best_model.predict(img_tensor, verbose=0)

    # ---- handle output model ----
    if isinstance(preds, dict):
        dr_key, dme_key = None, None
        for k in preds.keys():
            lk = k.lower()
            if "dr" in lk:
                dr_key = k
            if "dme" in lk:
                dme_key = k
        dr_pred = preds[dr_key][0]
        dme_pred = preds[dme_key][0]

    elif isinstance(preds, (list, tuple)):
        dr_pred = preds[0][0]
        dme_pred = preds[1][0]

    elif isinstance(preds, np.ndarray):
        dr_pred = preds[0][:5]
        dme_pred = preds[0][5:]

    else:
        raise ValueError("Format output model tidak dikenali")

    # ---- softmax ----
    dr_pred = ensure_probability(dr_pred)
    dme_pred = ensure_probability(dme_pred)

    # ---- result ----
    dr_idx = int(np.argmax(dr_pred))
    dme_idx = int(np.argmax(dme_pred))

    dr_name = DR_CLASSES[dr_idx]
    dme_name = DME_CLASSES[dme_idx]

    dr_conf = dr_pred[dr_idx] * 100
    dme_conf = dme_pred[dme_idx] * 100

    # ======================================================
    # 🔥 JUDUL UTAMA
    # ======================================================
    main_result = f"""
<div style="text-align:center; font-size:30px; font-weight:800; margin-bottom:20px;">
    HASIL PREDIKSI
</div>
"""

    # ======================================================
    # 📊 TABEL HASIL
    # ======================================================
    table_result = f"""
<table style="width:100%; border-collapse:collapse; font-size:16px;">
  <thead>
    <tr style="background-color:#f2f2f2;">
      <th style="border:1px solid #ccc; padding:8px;">Aspek</th>
      <th style="border:1px solid #ccc; padding:8px;">Klasifikasi</th>
      <th style="border:1px solid #ccc; padding:8px;">Confidence (%)</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td style="border:1px solid #ccc; padding:8px;">Diabetic Retinopathy (DR)</td>
      <td style="border:1px solid #ccc; padding:8px;"><b>{dr_name}</b></td>
      <td style="border:1px solid #ccc; padding:8px;">{dr_conf:.2f}%</td>
    </tr>
    <tr>
      <td style="border:1px solid #ccc; padding:8px;">Diabetic Macular Edema (DME)</td>
      <td style="border:1px solid #ccc; padding:8px;"><b>{dme_name}</b></td>
      <td style="border:1px solid #ccc; padding:8px;">{dme_conf:.2f}%</td>
    </tr>
  </tbody>
</table>
"""

    # ======================================================
    # 🩺 REKOMENDASI (DINAMIS)
    # ======================================================
    # Rekomendasi DR
    if dr_name in ["No DR"]:
        rec_dr = "Lanjutkan pola hidup sehat dan lakukan pemeriksaan mata rutin minimal 1 tahun sekali."
    elif dr_name in ["Mild", "Moderate"]:
        rec_dr = "Disarankan kontrol gula darah secara ketat dan pemeriksaan mata berkala setiap 6 bulan."
    else:  # Severe / Proliferative
        rec_dr = "Disarankan segera konsultasi ke dokter spesialis mata untuk evaluasi dan penanganan lebih lanjut."

    # Rekomendasi DME
    if dme_name == "No DME":
        rec_dme = "Belum ditemukan tanda edema makula diabetik, lanjutkan pemantauan rutin."
    elif dme_name == "Low Risk":
        rec_dme = "Perlu observasi ketat dan pemeriksaan lanjutan untuk mencegah progresivitas."
    else:  # High Risk
        rec_dme = "Disarankan segera mendapatkan evaluasi klinis dan terapi oleh dokter spesialis mata."

    recommendation = f"""
<div style="margin-top:20px; padding:15px; border-left:5px solid #007bff; background-color:#f9f9f9;">
<b>🩺 Rekomendasi Klinis:</b><br/><br/>
<b>• Diabetic Retinopathy (DR):</b><br/>
{rec_dr}<br/><br/>
<b>• Diabetic Macular Edema (DME):</b><br/>
{rec_dme}
</div>
"""

    return main_result + table_result + recommendation

# ============================================================
# 7. MULTI TEST IMAGES
# ============================================================
TEST_IMAGES = [
    "IDRiD_001test.jpg",
    "IDRiD_004test.jpg",
    "IDRiD_005test.jpg",
    "IDRiD_006test.jpg",
    "IDRiD_007test.jpg",
    "IDRiD_008test.jpg",
    "IDRiD_009test.jpg",
    "IDRiD_010test.jpg",
    "IDRiD_011test.jpg",
    "IDRiD_012test.jpg",
]

TEST_IMAGES = [[p] for p in TEST_IMAGES if os.path.exists(p)]

# ============================================================
# 8. GRADIO WRAPPER
# ============================================================
def gradio_predict(image):
    if image is None:
        return "❌ Silakan unggah gambar"

    temp_path = "uploaded.jpg"
    image.save(temp_path)

    return predict_and_format_text(temp_path)

# ============================================================
# 9. GRADIO UI
# ============================================================
with gr.Blocks(title="DR & DME Detection") as demo:

    gr.Markdown("""
    # 🩺 Diabetic Retinopathy Image Dataset (IDRiD)
    """)

    with gr.Row():
        with gr.Column():
            input_img = gr.Image(type="pil", label="Upload Gambar Retina")
            btn = gr.Button("🔍 Prediksi")

        with gr.Column():
            out_text = gr.Markdown()

    gr.Markdown("### 🧪 Data Testing")
    gr.Examples(
        examples=TEST_IMAGES,
        inputs=input_img
    )

    btn.click(
        fn=gradio_predict,
        inputs=input_img,
        outputs=out_text
    )

# ============================================================
# 10. LAUNCH
# ============================================================
demo.launch(share=True)