Delete app.py

app.py

@@ -1,113 +0,0 @@
-"""Gradio app for the FFDNet denoiser — roda em CPU no HF Spaces grátis."""
-from __future__ import annotations
-
-import os
-import urllib.request
-from pathlib import Path
-
-import gradio as gr
-import numpy as np
-import torch
-from PIL import Image
-
-from ffdnet_model import FFDNet
-
-WEIGHTS_DIR = Path(os.environ.get("FFDNET_WEIGHTS_DIR", "weights"))
-
-WEIGHTS = {
-    "color": {
-        "url": "https://github.com/cszn/KAIR/releases/download/v1.0/ffdnet_color.pth",
-        "file": WEIGHTS_DIR / "ffdnet_color.pth",
-        "in_nc": 3,
-        "out_nc": 3,
-        "nc": 96,
-        "nb": 12,
-    },
-    "gray": {
-        "url": "https://github.com/cszn/KAIR/releases/download/v1.0/ffdnet_gray.pth",
-        "file": WEIGHTS_DIR / "ffdnet_gray.pth",
-        "in_nc": 1,
-        "out_nc": 1,
-        "nc": 64,
-        "nb": 15,
-    },
-}
-
-DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
-_MODELS: dict[str, FFDNet] = {}
-
-
-def _download(url: str, target: Path) -> None:
-    if target.exists():
-        return
-    target.parent.mkdir(parents=True, exist_ok=True)
-    print(f"[ffdnet] downloading {url} -> {target}")
-    urllib.request.urlretrieve(url, target)
-
-
-def _load(mode: str) -> FFDNet:
-    if mode in _MODELS:
-        return _MODELS[mode]
-    cfg = WEIGHTS[mode]
-    _download(cfg["url"], cfg["file"])
-    model = FFDNet(in_nc=cfg["in_nc"], out_nc=cfg["out_nc"], nc=cfg["nc"], nb=cfg["nb"])
-    state = torch.load(cfg["file"], map_location=DEVICE, weights_only=True)
-    # Strip DataParallel prefix se existir
-    state = {k.replace("module.", "", 1): v for k, v in state.items()}
-    model.load_state_dict(state, strict=True)
-    model.eval().to(DEVICE)
-    _MODELS[mode] = model
-    return model
-
-
-def denoise(image: Image.Image, noise_sigma: float, mode: str) -> Image.Image:
-    if image is None:
-        raise gr.Error("Envie uma imagem.")
-    if mode not in WEIGHTS:
-        raise gr.Error(f"Modo inválido: {mode}")
-
-    model = _load(mode)
-
-    pil = image.convert("RGB" if mode == "color" else "L")
-    arr = np.array(pil)
-    if arr.ndim == 2:
-        arr = arr[..., None]
-
-    tensor = (
-        torch.from_numpy(arr.astype(np.float32) / 255.0)
-        .permute(2, 0, 1)
-        .unsqueeze(0)
-        .to(DEVICE)
-    )
-    sigma = torch.tensor([float(noise_sigma) / 255.0], dtype=torch.float32, device=DEVICE)
-
-    with torch.no_grad():
-        out = model(tensor, sigma).clamp(0.0, 1.0)
-
-    out_np = (out.squeeze(0).permute(1, 2, 0).cpu().numpy() * 255.0).round().astype(np.uint8)
-    if mode == "gray":
-        out_np = out_np[..., 0]
-    return Image.fromarray(out_np)
-
-
-with gr.Blocks(title="FFDNet — Denoiser") as demo:
-    gr.Markdown(
-        "# FFDNet — Denoiser\n"
-        "CNN denoiser com nível de ruído (σ) ajustável. "
-        "Útil para limpar scans/fotos antes de OCR. "
-        "Primeira chamada baixa os pesos (~4MB/10MB) e pode demorar alguns segundos."
-    )
-    with gr.Row():
-        with gr.Column():
-            inp = gr.Image(type="pil", label="Imagem de entrada")
-            sigma = gr.Slider(0, 75, value=15, step=1, label="σ — nível de ruído estimado")
-            mode = gr.Radio(choices=["color", "gray"], value="color", label="Modo")
-            btn = gr.Button("Denoise", variant="primary")
-        with gr.Column():
-            out_img = gr.Image(type="pil", label="Resultado")
-
-    btn.click(fn=denoise, inputs=[inp, sigma, mode], outputs=out_img, api_name="denoise")
-
-
-if __name__ == "__main__":
-    demo.queue(max_size=8).launch()