Create app.py

app.py

@@ -0,0 +1,112 @@
+import gradio as gr
+import torch
+import cv2
+import numpy as np
+from PIL import Image
+import base64
+import io
+from realesrgan import RealESRGANer
+from basicsr.archs.rrdbnet_arch import RRDBNet
+
+# === KONFIGURASI MODEL ===
+# Kita menggunakan RealESRGAN_x4plus (bagus untuk general image/realistis)
+# Jika kamu lebih fokus ke anime, bisa ganti ke RealESRGAN_x4plus_anime_6B
+MODEL_NAME = 'RealESRGAN_x4plus'
+MODEL_PATH = 'https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth'
+
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+print(f"Menggunakan device: {device}")
+
+# Inisialisasi arsitektur model
+model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=4)
+
+# Inisialisasi RealESRGANer
+upsampler = RealESRGANer(
+    scale=4,
+    model_path=MODEL_PATH,
+    model=model,
+    tile=0, # Atur > 0 jika kehabisan memori (misal: 400), 0 berarti tidak di-tile (lebih cepat tapi butuh VRAM besar)
+    tile_pad=10,
+    pre_pad=0,
+    half=True if torch.cuda.is_available() else False, # Gunakan presisi half jika pakai GPU agar lebih cepat
+    device=device
+)
+
+# === FUNGSI UTAMA ===
+def decode_base64_to_cv2(base64_string):
+    """Mengubah base64 dari web menjadi format gambar OpenCV (numpy array)"""
+    # Pisahkan header "data:image/png;base64," jika ada
+    if "," in base64_string:
+        base64_string = base64_string.split(",")[1]
+    
+    img_data = base64.b64decode(base64_string)
+    nparr = np.frombuffer(img_data, np.uint8)
+    img_cv2 = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
+    return img_cv2
+
+def encode_cv2_to_base64(img_cv2, ext=".png"):
+    """Mengubah format gambar OpenCV kembali ke base64 untuk dikirim ke web"""
+    _, buffer = cv2.imencode(ext, img_cv2)
+    b64_str = base64.b64encode(buffer).decode("utf-8")
+    return f"data:image/png;base64,{b64_str}"
+
+def upscale_image(base64_image, target_res):
+    """
+    Fungsi utama yang akan dipanggil oleh API.
+    target_res: '2k' atau '4k'
+    """
+    try:
+        # 1. Decode gambar
+        img = decode_base64_to_cv2(base64_image)
+        
+        # 2. Lakukan Upscale (default model ini menaikkan 4x lipat)
+        print("Memulai proses upscaling...")
+        output_img, _ = upsampler.enhance(img, outscale=4)
+        
+        # 3. Resize cerdas (Opsional)
+        # RealESRGAN otomatis x4. Jika hasilnya kebesaran, kita resize manual ke target yang diinginkan.
+        h, w = output_img.shape[:2]
+        
+        target_w, target_h = w, h
+        if target_res.lower() == '2k':
+            # Asumsi 2K lebar maksimal ~2560px
+            max_size = 2560
+            if max(w, h) > max_size:
+                scale_ratio = max_size / max(w, h)
+                target_w = int(w * scale_ratio)
+                target_h = int(h * scale_ratio)
+                output_img = cv2.resize(output_img, (target_w, target_h), interpolation=cv2.INTER_AREA)
+        elif target_res.lower() == '4k':
+            # Asumsi 4K lebar maksimal ~3840px
+            max_size = 3840
+            if max(w, h) > max_size:
+                scale_ratio = max_size / max(w, h)
+                target_w = int(w * scale_ratio)
+                target_h = int(h * scale_ratio)
+                output_img = cv2.resize(output_img, (target_w, target_h), interpolation=cv2.INTER_AREA)
+
+        print(f"Upscale selesai. Resolusi akhir: {target_w}x{target_h}")
+        
+        # 4. Encode kembali ke base64
+        result_base64 = encode_cv2_to_base64(output_img)
+        return result_base64
+        
+    except Exception as e:
+        print(f"Error saat upscaling: {e}")
+        return str(e)
+
+# === ANTARMUKA API GRADIO ===
+# Kita definisikan interface tanpa UI web yang rumit karena kita hanya butuh API-nya
+with gr.Blocks() as demo:
+    with gr.Row():
+        input_b64 = gr.Textbox(label="Input Base64 Image")
+        input_res = gr.Textbox(label="Target Resolution (2k/4k)", value="2k")
+    
+    output_b64 = gr.Textbox(label="Output Base64 Image")
+    
+    btn = gr.Button("Upscale")
+    btn.click(fn=upscale_image, inputs=[input_b64, input_res], outputs=output_b64, api_name="predict")
+
+# Jalankan server
+if __name__ == "__main__":
+    demo.queue().launch()