Update app.py

app.py

@@ -1,68 +1,191 @@
 import os
-import tempfile
-import gradio as gr
-from rembg import remove
+import math
+import uuid
+import numpy as np
+import onnxruntime as ort
 from PIL import Image
+import gradio as gr
+import tempfile
+import requests
+
+# ——————————————————————————————————————————————————————————————
+# 1) Path ke ONNX model “×4” (diunduh dari GitHub)
+MODEL_DIR     = "model"
+MODEL_X4_PATH = os.path.join(MODEL_DIR, "Real-ESRGAN-x4plus.onnx")
+
+# Cek jika model belum ada, maka unduh dari GitHub
+if not os.path.isfile(MODEL_X4_PATH):
+    os.makedirs(MODEL_DIR, exist_ok=True)
+    model_url = "https://raw.githubusercontent.com/salmanalfarisi11/Upscaler_images/master/model/Real-ESRGAN-x4plus.onnx"
+    response = requests.get(model_url)
+    
+    # Jika unduhan berhasil, simpan model
+    if response.status_code == 200:
+        with open(MODEL_X4_PATH, 'wb') as f:
+            f.write(response.content)
+        print(f"Model telah diunduh dan disimpan di {MODEL_X4_PATH}")
+    else:
+        raise FileNotFoundError(f"Model gagal diunduh dari GitHub, status code: {response.status_code}")
+
+# ——————————————————————————————————————————————————————————————
+# 2) Buat ONNXRuntime session hanya dengan CPU, 2 thread
+sess_opts = ort.SessionOptions()
+sess_opts.intra_op_num_threads = 2
+sess_opts.inter_op_num_threads = 2
+
+session_x4 = ort.InferenceSession(MODEL_X4_PATH, sess_options=sess_opts,
+                                  providers=["CPUExecutionProvider"])
+
+# Ambil metadata ukuran input tile untuk model ×4 (biasanya 128×128)
+input_meta_x4 = session_x4.get_inputs()[0]
+_, _, H_in_x4, W_in_x4 = tuple(input_meta_x4.shape)
+H_in_x4, W_in_x4 = int(H_in_x4), int(W_in_x4)
 
-# Force ONNXRuntime to CPU only
-os.environ["ORT_PROVIDERS"] = "CPUExecutionProvider"
+# Cek skala (harusnya 4×)
+dummy = np.zeros((1, 3, H_in_x4, W_in_x4), dtype=np.float32)
+dummy_out = session_x4.run(None, {input_meta_x4.name: dummy})[0]
+_, _, H_out_x4, W_out_x4 = dummy_out.shape
+SCALE_X4 = H_out_x4 // H_in_x4
+if SCALE_X4 != 4:
+    raise RuntimeError(f"Model ×4 menghasilkan scale = {SCALE_X4}, bukan 4")
 
-def remove_background(input_image: Image.Image) -> Image.Image:
-    """Your existing remover."""
-    return remove(input_image)
+# ——————————————————————————————————————————————————————————————
+# 3) Fungsi util untuk mem‐proses satu tile 128×128 → 512×512
+def run_tile_x4(tile_np: np.ndarray) -> np.ndarray:
+    """
+    Input: tile_np (128,128,3) float32 ∈ [0,1]
+    Output: (512,512,3) float32 ∈ [0,1]
+    """
+    patch_nchw = np.transpose(tile_np, (2, 0, 1))[None, ...]  # → (1,3,128,128)
+    out_nchw = session_x4.run(None, {input_meta_x4.name: patch_nchw})[0]  # (1,3,512,512)
+    out_nchw = np.squeeze(out_nchw, axis=0)            # (3,512,512)
+    out_hwc  = np.transpose(out_nchw, (1, 2, 0))        # (512,512,3)
+    return out_hwc  # float32 ∈ [0,1]
 
-def remove_and_save(img: Image.Image):
+# ——————————————————————————————————————————————————————————————
+# 4) Pipeline tile‐based untuk ×4 –– mem‐pad, potong, infer, reassemble, crop
+def tile_upscale_x4(input_img: Image.Image):
     """
-    1) remove background,
-    2) save to a temp PNG,
-    3) return (PIL.Image for preview, file path for DownloadButton).
+    1. PIL→float32 np ∈ [0,1]
+    2. Pad agar (H,W) kelipatan H_in_x4 (128)
+    3. Bagi menjadi tile (H_in_x4 × W_in_x4)
+    4. run_tile_x4 per tile, simpan ke out_arr
+    5. Crop ke (orig_h*4, orig_w*4)
+    6. Convert ke uint8 PIL, save PNG, return (PIL, filepath)
     """
-    result = remove_background(img)
+    # 4.1. Convert PIL→float32 [0,1]
+    img_rgb = input_img.convert("RGB")
+    arr = np.array(img_rgb).astype(np.float32) / 255.0  # (h_orig, w_orig, 3)
+    h_orig, w_orig, _ = arr.shape
+
+    # 4.2. Hitung jumlah tile dan padding
+    tiles_h = math.ceil(h_orig / H_in_x4)
+    tiles_w = math.ceil(w_orig / W_in_x4)
+    pad_h   = tiles_h * H_in_x4 - h_orig
+    pad_w   = tiles_w * W_in_x4 - w_orig
+
+    # Reflect pad di kanan + bawah
+    arr_padded = np.pad(
+        arr,
+        ((0, pad_h), (0, pad_w), (0, 0)),
+        mode="reflect"
+    )  # → (tiles_h * 128, tiles_w * 128, 3)
+
+    # 4.3. Buat array output berukuran (tiles_h*512, tiles_w*512, 3)
+    out_h = tiles_h * H_in_x4 * SCALE_X4
+    out_w = tiles_w * W_in_x4 * SCALE_X4
+    out_arr = np.zeros((out_h, out_w, 3), dtype=np.float32)
+
+    # 4.4. Loop semua tile
+    for i in range(tiles_h):
+        for j in range(tiles_w):
+            y0   = i * H_in_x4
+            x0   = j * W_in_x4
+            tile = arr_padded[y0 : y0 + H_in_x4, x0 : x0 + W_in_x4, :]  # (128,128,3)
+
+            up_tile = run_tile_x4(tile)  # (512,512,3) float32
+
+            oy0 = i * H_in_x4 * SCALE_X4
+            ox0 = j * W_in_x4 * SCALE_X4
+            out_arr[oy0 : oy0 + H_in_x4 * SCALE_X4,
+                    ox0 : ox0 + W_in_x4 * SCALE_X4, :] = up_tile
+
+    # 4.5. Crop → (h_orig*4, w_orig*4)
+    final_arr = out_arr[0 : h_orig * SCALE_X4, 0 : w_orig * SCALE_X4, :]
+    final_arr = np.clip(final_arr, 0.0, 1.0)
+    final_uint8 = (final_arr * 255.0).round().astype(np.uint8)
+    final_pil   = Image.fromarray(final_uint8)  # (h_orig*4, w_orig*4)
+
+    # 4.6. Simpan ke file PNG unik
     tmp = tempfile.NamedTemporaryFile(delete=False, suffix=".png")
-    result.save(tmp.name, format="PNG")
+    final_pil.save(tmp.name, format="PNG")
     tmp.close()
-    return result, tmp.name
+    return final_pil, tmp.name
+
+# ——————————————————————————————————————————————————————————————
+# 5) Fungsi Standard (×4) dan Premium (×8) untuk Gradio
+def standard_upscale(input_img: Image.Image):
+    # Hasil ×4 tile→reassemble 
+    return tile_upscale_x4(input_img)
+
+def premium_upscale(input_img: Image.Image):
+    """
+    - Jalankan pipeline ×4 untuk dapat final_4x (PIL)
+    - Lalu bicubic‐resize final_4x → 8× resolusi (orig_w*8, orig_h*8)
+    - Simpan PNG baru dan return (PIL_8x, filepath)
+    """
+    # 5.1. Pertama dapatkan output ×4
+    final_4x, path_4x = tile_upscale_x4(input_img)  # final_4x size = (h_orig*4, w_orig*4)
 
-with gr.Blocks() as demo:
-    gr.Markdown("## 🖼️ Background Remover")
+    # 5.2. Ukuran asli
+    w_orig, h_orig = input_img.size
+
+    # 5.3. Resize bicubic (LANCZOS) dari 4× → 8×
+    target_size = (w_orig * 8, h_orig * 8)
+    final_8x = final_4x.resize(target_size, resample=Image.LANCZOS)
+
+    # 5.4. Simpan PNG unik
+    tmp = tempfile.NamedTemporaryFile(delete=False, suffix=".png")
+    final_8x.save(tmp.name, format="PNG")
+    tmp.close()
+
+    return final_8x, tmp.name
+
+# ——————————————————————————————————————————————————————————————
+# 6) CSS Kustom agar tombol Premium bergenre “emas”
+css = """
+#premium-btn {
+    background-color: gold !important;
+    color: black !important;
+}
+"""
+
+# ——————————————————————————————————————————————————————————————
+# 7) Bangun Gradio Blocks – 2 tombol berdampingan
+with gr.Blocks(css=css, title="Real-ESRGAN Dual-Mode Upscaler") as demo:
     gr.Markdown(
-        "Upload an image and click **Remove Background** to get a transparent PNG.  \n"
-        "**Note:** CPU-only ONNX Runtime (no GPU needed)."
+        """
+        # Real-ESRGAN Dual-Mode Upscaler  
+        **Standard Upscale** (×4) atau **Premium Upscale 🚀** (×8).  
+        Silakan upload gambar apa saja, kemudian klik tombol yang diinginkan.
+        """
     )
 
+    # Baris untuk upload gambar
     with gr.Row():
-        original = gr.Image(type="pil", label="Original")
-        result   = gr.Image(type="pil", label="Result")
-
-    # Stack Remove + Download vertically
-    with gr.Column(elem_id="controls"):
-        remove_btn   = gr.Button("✂️ Remove Background", variant="primary")
-        download_btn = gr.DownloadButton("⬇️ Download PNG")
-
-    remove_btn.click(
-        fn=remove_and_save,
-        inputs=original,
-        outputs=[result, download_btn],
-    )
+        inp_image = gr.Image(type="pil", label="Upload Source Image")
 
-    # Centered footer
-    demo.css = """
-      #controls { align-items: start; gap: 0.5rem; }
-      .footer { text-align: center; margin-top: 2rem; color: #666; font-size:0.9rem; }
-      .social-icons a { margin: 0 0.5rem; text-decoration: none; color: inherit; }
-    """
+    # Baris untuk dua tombol
+    with gr.Row():
+        btn_std  = gr.Button("Standard Upscale (×4)", variant="primary", elem_id="std-btn")
+        btn_prem = gr.Button("Premium Upscale 🚀 (×8)", elem_id="premium-btn")
+
+    # Dua output: preview image & link “Download PNG”
+    out_preview  = gr.Image(type="pil", label="Upscaled Preview")
+    out_download = gr.DownloadButton("⬇️ Download PNG", visible=True)
+
+    # Hubungkan tombol ke fungsi:
+    btn_std.click(fn=standard_upscale, inputs=inp_image, outputs=[out_preview, out_download])
+    btn_prem.click(fn=premium_upscale, inputs=inp_image, outputs=[out_preview, out_download])
 
-    gr.HTML("""
-      <div class="footer">
-        Made by Salman Alfarisi • 2025<br>
-        <span class="social-icons">
-          <a href="https://github.com/salmanalfarisi11" target="_blank">GitHub</a> |
-          <a href="https://www.instagram.com/faris.salman111/" target="_blank">Instagram</a> |
-          <a href="https://id.linkedin.com/in/salmanalfarisi11" target="_blank">LinkedIn</a>
-        </span><br>
-        Source code on <a href="https://github.com/salmanalfarisi11" target="_blank">GitHub</a>
-      </div>
-    """)
-
-if __name__ == "__main__":
-    demo.launch()
+demo.launch(server_name="0.0.0.0", server_port=7860)