Update app.py

app.py

@@ -1,25 +1,141 @@
+import os
+import time
+import torch
+import warnings
+import numpy as np
 import gradio as gr
+import matplotlib.pyplot as plt
 from PIL import Image
-from model import predict
+from sklearn.metrics import average_precision_score
+from model import (
+    load_classification_model,
+    load_t5_model,
+    predict_image,
+    generate_comment_turkce,
+    calculate_performance_metrics
+)
+
+warnings.filterwarnings("ignore", category=FutureWarning)
+warnings.filterwarnings("ignore", category=UserWarning)
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+cnn_model_f = load_classification_model(device, model_type="f", num_classes=6)
+cnn_model_c = load_classification_model(device, model_type="c", num_classes=6)
+cnn_model_q = load_classification_model(device, model_type="q", num_classes=6)
+
+t5_tokenizer, t5_model = load_t5_model(device)
+
+perf_metrics_f = calculate_performance_metrics(cnn_model_f, device)
+perf_metrics_c = calculate_performance_metrics(cnn_model_c, device)
+perf_metrics_q = calculate_performance_metrics(cnn_model_q, device)
+
+class_names_en = [
+    "Alzheimer Disease",
+    "Mild Alzheimer Risk",
+    "Moderate Alzheimer Risk",
+    "Very Mild Alzheimer Risk",
+    "No Risk",
+    "Parkinson Disease"
+]
+en2tr = {
+    "Alzheimer Disease": "Alzheimer Hastalığı",
+    "Mild Alzheimer Risk": "Hafif Alzheimer Riski",
+    "Moderate Alzheimer Risk": "Orta Düzey Alzheimer Riski",
+    "Very Mild Alzheimer Risk": "Çok Hafif Alzheimer Riski",
+    "No Risk": "Risk Yok",
+    "Parkinson Disease": "Parkinson Hastalığı"
+}
+
+def gradio_predict(image, model_type, question):
+    if model_type == "f":
+        cnn_model = cnn_model_f
+        perf_metrics = perf_metrics_f
+    elif model_type == "c":
+        cnn_model = cnn_model_c
+        perf_metrics = perf_metrics_c
+    elif model_type == "q":
+        cnn_model = cnn_model_q
+        perf_metrics = perf_metrics_q
+    else:
+        return None, "Hata: Geçersiz model tipi", "", ""
+
+    idx, conf, inp_tensor, all_probs = predict_image(cnn_model, image, device)
+    pred_en = class_names_en[idx]
+    pred_tr = en2tr[pred_en]
+
+    if question is None or question.strip() == "":
+        comment = generate_comment_turkce(t5_tokenizer, t5_model, pred_tr, device)
+    else:
+        input_text = f"Sınıf: {pred_tr}. Soru: {question}"
+        inputs = t5_tokenizer(
+            input_text,
+            return_tensors="pt",
+            padding="longest",
+            truncation=True,
+            max_length=64
+        ).to(device)
+        out_ids = t5_model.generate(
+            **inputs,
+            max_length=64,
+            do_sample=True,
+            top_k=50,
+            top_p=0.95,
+            no_repeat_ngram_size=2,
+            early_stopping=True
+        )
+        comment = t5_tokenizer.decode(out_ids[0], skip_special_tokens=True)
+
+    num_classes = len(class_names_en)
+    true_onehot = np.zeros(num_classes, dtype=int)
+    true_onehot[idx] = 1
+    ap_scores = []
+    for cls_i in range(num_classes):
+        true_binary = (true_onehot == 1).astype(int) if cls_i == idx else (true_onehot == 0).astype(int)
+        score = all_probs[cls_i]
+        ap = average_precision_score(true_binary, [score])
+        ap_scores.append(ap)
+    map_score = np.mean(ap_scores)
+
+    inp_np = inp_tensor.squeeze(0).permute(1, 2, 0).cpu().numpy()
+    mean = np.array([0.485, 0.456, 0.406])
+    std = np.array([0.229, 0.224, 0.225])
+    img_show = inp_np * std + mean
+    img_show = np.clip(img_show, 0, 1)
+
+    tahmin_metni = f"Tahmin: {pred_en} — {conf:.2f}%"
+    ap_map_metni = f"AP (sınıf {pred_en}): {ap_scores[idx]:.4f}  |  mAP: {map_score:.4f}"
+
+    return img_show, tahmin_metni, ap_map_metni, comment
 
-def interface_fn(image, question):
-    prediction, confidence, comment = predict(image, question)
-    return f"Tahmin: {prediction} ({confidence*100:.2f}%)", comment
+css = """
+img {border-radius: 8px;}
+"""
 
 demo = gr.Interface(
-    fn=interface_fn,
+    fn=gradio_predict,
     inputs=[
         gr.Image(type="pil", label="MRI Görüntüsü Yükle"),
-        gr.Textbox(label="Görselle ilgili soru", placeholder="Örnek: Bu hasta tedaviye ihtiyaç duyar mı?")
+        gr.Radio(choices=["f", "c", "q"], label="Vbai Model Tipi Seçin"),
+        gr.Textbox(label="Görselle ilgili soru (isteğe bağlı)", placeholder="Örnek: Bu hasta tedaviye ihtiyaç duyar mı?")
     ],
     outputs=[
-        gr.Textbox(label="Tahmin Sonucu"),
+        gr.Image(type="numpy", label="Ön İşlenmiş Görsel"),
+        gr.Textbox(label="Tahmin ve Güven"),
+        gr.Textbox(label="AP ve mAP"),
         gr.Textbox(label="Tbai Yorumu")
     ],
-    title="Vbai-DPA 2.3c & Tbai-DPA 1.0-BETA🧠",
-    description="MRI görüntüsü ile hastalık sınıflandırması yapar ve Tbai modeli ile tıbbi yorum üretir.",
-    theme="soft"
+    title="🧠 Vbai-DPA 2.3 | Seçilebilir Vbai (f/c/q) + Tbai Yorumu",
+    description=(
+        "1) MRI görüntüsünü yükleyin.\n"
+        "2) Vbai için f, c veya q modellerinden birini seçin.\n"
+        "3) Opsiyonel olarak görselle ilgili kısa bir soru girin.\n"
+        "4) Tahmin sonucu, gerçek AP/mAP ve Tbai tabanlı Türkçe yorum ekranda görülecek.\n"
+        "5) NOT: Sonuçlar yapay zeka tarafından üretildiğinden, doktorunuzdan en doğru bilgiye ulaşmanız gereklidir!"
+    ),
+    theme="default",
+    css=css
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()