app.py

# ────────────────────────────────────────────────────────
# TorchVision compat shim (MUST be before importing basicsr)
# Fixes: ModuleNotFoundError: torchvision.transforms.functional_tensor
# ────────────────────────────────────────────────────────
import sys, types
try:
    import torchvision.transforms.functional_tensor as _ft  # noqa: F401
except Exception:
    from torchvision.transforms import functional as _F
    _mod = types.ModuleType("torchvision.transforms.functional_tensor")
    _mod.rgb_to_grayscale = _F.rgb_to_grayscale
    sys.modules["torchvision.transforms.functional_tensor"] = _mod

# ────────────────────────────────────────────────────────
# Spaces ZeroGPU decorator (safe no-op locally)
# ────────────────────────────────────────────────────────
try:
    import spaces
    GPU = spaces.GPU
except Exception:
    def GPU(*args, **kwargs):
        def _wrap(f): return f
        return _wrap

# ────────────────────────────────────────────────────────
# Standard imports
# ────────────────────────────────────────────────────────
import gradio as gr
import cv2
import numpy
import os
import random
import inspect
from pathlib import Path
import zipfile
import tempfile


from basicsr.archs.rrdbnet_arch import RRDBNet as _RRDBNet
from basicsr.utils.download_util import load_file_from_url

from realesrgan import RealESRGANer
from realesrgan.archs.srvgg_arch import SRVGGNetCompact

# ────────────────────────────────────────────────────────
# Globals
# ────────────────────────────────────────────────────────
last_file = None
img_mode = "RGBA"


# ────────────────────────────────────────────────────────
# Utilities
# ────────────────────────────────────────────────────────
def rnd_string(x: int) -> str:
    characters = "abcdefghijklmnopqrstuvwxyz_0123456789"
    return "".join((random.choice(characters)) for _ in range(x))


def reset():
    global last_file
    if last_file:
        try:
            print(f"Deleting {last_file} ...")
            os.remove(last_file)
        except Exception as e:
            print("Delete error:", e)
        last_file = None
    return gr.update(value=None), gr.update(value=None)


def has_transparency(img):
    if img.info.get("transparency", None) is not None:
        return True
    if img.mode == "P":
        transparent = img.info.get("transparency", -1)
        for _, index in img.getcolors():
            if index == transparent:
                return True
    elif img.mode == "RGBA":
        extrema = img.getextrema()
        if extrema[3][0] < 255:
            return True
    return False


def image_properties(img):
    global img_mode
    if img:
        img_mode = "RGBA" if has_transparency(img) else "RGB"
        return f"Resolution: Width: {img.size[0]}, Height: {img.size[1]}  |  Color Mode: {img_mode}"


def model_tip_text(model_name: str) -> str:
    tips = {
        "RealESRGAN_x4plus": (
            "**RealESRGAN_x4plus (4×)** — Best for photoreal images (portraits, landscapes). "
            "Balanced detail recovery. Good default for Flux realism."
        ),
        "RealESRNet_x4plus": (
            "**RealESRNet_x4plus (4×)** — Softer but great on noisy/compressed sources "
            "(old JPEGs, screenshots)."
        ),
        "RealESRGAN_x4plus_anime_6B": (
            "**RealESRGAN_x4plus_anime_6B (4×)** — For anime/illustrations/line art only. "
            "Not recommended for real-life photos."
        ),
        "RealESRGAN_x2plus": (
            "**RealESRGAN_x2plus (2×)** — Faster, lighter 2× cleanup when you don't need 4×."
        ),
        "realesr-general-x4v3": (
            "**realesr-general-x4v3 (4×)** — Versatile mixed-content model with adjustable denoise. "
            "**Denoise Strength** slider only affects this model (blends with the WDN variant). "
            "Try 0.3–0.5 for slightly cleaner, sharper results."
        ),
    }
    return tips.get(model_name, "")


# ────────────────────────────────────────────────────────
# RRDBNet builder that tolerates different Basicsr signatures
# ────────────────────────────────────────────────────────
def build_rrdb(scale: int, num_block: int):
    """
    Creates an RRDBNet across several possible constructor signatures used by basicsr/realesrgan.
    Tries, in order:
      1) keyword style (num_in_ch/num_out_ch/num_feat/num_block/num_grow_ch/scale)
      2) alt keyword style (in_nc/out_nc/nf/nb/gc/sf)
      3) positional with gc before scale
      4) positional with scale before gc
    """
    # Try keyword: "num_*" + "scale"
    try:
        return _RRDBNet(
            num_in_ch=3, num_out_ch=3,
            num_feat=64, num_block=num_block,
            num_grow_ch=32, scale=scale
        )
    except TypeError:
        pass

    # Try keyword: "in_nc/out_nc" + "sf"
    try:
        return _RRDBNet(
            in_nc=3, out_nc=3,
            nf=64, nb=num_block,
            gc=32, sf=scale
        )
    except TypeError:
        pass

    # Inspect parameters to guess positional order
    params = list(inspect.signature(_RRDBNet).parameters.keys())

    # Common positional (gc, scale) order
    try:
        return _RRDBNet(3, 3, 64, num_block, 32, scale)
    except TypeError:
        pass

    # Alternate positional (scale, gc) order
    try:
        return _RRDBNet(3, 3, 64, num_block, scale, 32)
    except TypeError as e:
        raise TypeError(f"RRDBNet signature not recognized: {e}")

#Factor an upsampler builder
def get_upsampler(model_name: str, outscale: int, tile: int = 256):
    # Build the same backbone/weights as in realesrgan(), but return a ready RealESRGANer
    if model_name == 'RealESRGAN_x4plus':
        model = build_rrdb(scale=4, num_block=23); netscale = 4
        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth']
    elif model_name == 'RealESRNet_x4plus':
        model = build_rrdb(scale=4, num_block=23); netscale = 4
        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.1/RealESRNet_x4plus.pth']
    elif model_name == 'RealESRGAN_x4plus_anime_6B':
        model = build_rrdb(scale=4, num_block=6); netscale = 4
        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth']
    elif model_name == 'RealESRGAN_x2plus':
        model = build_rrdb(scale=2, num_block=23); netscale = 2
        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.1/RealESRGAN_x2plus.pth']
    elif model_name == 'realesr-general-x4v3':
        model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=32, upscale=4, act_type='prelu'); netscale = 4
        file_url = [
            'https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth',
            'https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-wdn-x4v3.pth'
        ]
    else:
        raise ValueError(f"Unknown model: {model_name}")

    ROOT_DIR = os.path.dirname(os.path.abspath(__file__))
    weights_dir = os.path.join(ROOT_DIR, 'weights')
    os.makedirs(weights_dir, exist_ok=True)
    for url in file_url:
        fname = os.path.basename(url)
        local_path = os.path.join(weights_dir, fname)
        if not os.path.isfile(local_path):
            load_file_from_url(url=url, model_dir=weights_dir, progress=True)

    if model_name == 'realesr-general-x4v3':
        model_path = [
            os.path.join(weights_dir, 'realesr-general-x4v3.pth'),
            os.path.join(weights_dir, 'realesr-general-wdn-x4v3.pth'),
        ]
        dni_weight = None  # supplied at call site if using denoise blend
    else:
        model_path = os.path.join(weights_dir, f"{model_name}.pth")
        dni_weight = None

    use_cuda = False
    try:
        use_cuda = hasattr(cv2, "cuda") and cv2.cuda.getCudaEnabledDeviceCount() > 0
    except Exception:
        use_cuda = False
    gpu_id = 0 if use_cuda else None

    upsampler = RealESRGANer(
        scale=netscale,
        model_path=model_path,
        dni_weight=dni_weight,
        model=model,
        tile=tile or 256,
        tile_pad=10,
        pre_pad=10,
        half=bool(use_cuda),
        gpu_id=gpu_id
    )
    return upsampler, netscale, use_cuda, model_path


# ────────────────────────────────────────────────────────
# Core upscaling
# Decorated for Hugging Face Spaces ZeroGPU
# ────────────────────────────────────────────────────────
@GPU()  # lets Spaces know this function uses GPU; safe no-op locally
def realesrgan(img, model_name, denoise_strength, face_enhance, outscale):
    if img is None:
        return

    # ----- Select backbone + weights -----
    if model_name == 'RealESRGAN_x4plus':
        model = build_rrdb(scale=4, num_block=23); netscale = 4
        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth']

    elif model_name == 'RealESRNet_x4plus':
        model = build_rrdb(scale=4, num_block=23); netscale = 4
        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.1/RealESRNet_x4plus.pth']

    elif model_name == 'RealESRGAN_x4plus_anime_6B':
        model = build_rrdb(scale=4, num_block=6); netscale = 4
        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth']

    elif model_name == 'RealESRGAN_x2plus':
        model = build_rrdb(scale=2, num_block=23); netscale = 2
        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.1/RealESRGAN_x2plus.pth']

    elif model_name == 'realesr-general-x4v3':
        model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=32, upscale=4, act_type='prelu'); netscale = 4
        file_url = [
            'https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth',
            'https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-wdn-x4v3.pth'
        ]
    else:
        raise ValueError(f"Unknown model: {model_name}")

    # ----- Ensure weights on disk -----
    ROOT_DIR = os.path.dirname(os.path.abspath(__file__))
    weights_dir = os.path.join(ROOT_DIR, 'weights')
    os.makedirs(weights_dir, exist_ok=True)

    for url in file_url:
        fname = os.path.basename(url)
        local_path = os.path.join(weights_dir, fname)
        if not os.path.isfile(local_path):
            load_file_from_url(url=url, model_dir=weights_dir, progress=True)

    if model_name == 'realesr-general-x4v3':
        base_path = os.path.join(weights_dir, 'realesr-general-x4v3.pth')
        wdn_path  = os.path.join(weights_dir, 'realesr-general-wdn-x4v3.pth')
        model_path = [base_path, wdn_path]
        denoise_strength = float(denoise_strength)
        dni_weight = [1.0 - denoise_strength, denoise_strength]  # base, WDN
    else:
        model_path = os.path.join(weights_dir, f"{model_name}.pth")
        dni_weight = None

    # ----- CUDA / precision / tiling -----
    use_cuda = False
    try:
        import torch
        use_cuda = torch.cuda.is_available()
    except Exception:
        use_cuda = False
    gpu_id = 0 if use_cuda else None

    upsampler = RealESRGANer(
        scale=netscale,
        model_path=model_path,
        dni_weight=dni_weight,
        model=model,
        tile=256,       # VRAM-safe default; lower to 128 if OOM
        tile_pad=10,
        pre_pad=10,
        half=bool(use_cuda),
        gpu_id=gpu_id
    )

    # ----- Optional face enhancement -----
    face_enhancer = None
    if face_enhance:
        from gfpgan import GFPGANer
        face_enhancer = GFPGANer(
            model_path='https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth',
            upscale=outscale,
            arch='clean',
            channel_multiplier=2,
            bg_upsampler=upsampler
        )

    # ----- PIL -> cv2 -----
    cv_img = numpy.array(img)
    if cv_img.ndim == 3 and cv_img.shape[2] == 4:
        cv_img = cv2.cvtColor(cv_img, cv2.COLOR_RGBA2BGRA)
    else:
        cv_img = cv2.cvtColor(cv_img, cv2.COLOR_RGB2BGR)

    # ----- Enhance -----
    try:
        if face_enhancer:
            _, _, output = face_enhancer.enhance(cv_img, has_aligned=False, only_center_face=False, paste_back=True)
        else:
            output, _ = upsampler.enhance(cv_img, outscale=int(outscale))
    except RuntimeError as error:
        print('Error', error)
        print('Tip: If you hit CUDA OOM, try a smaller tile size (e.g., 128).')
        return None

    # ----- cv2 -> display ndarray, also save -----
    if output.ndim == 3 and output.shape[2] == 4:
        display_img = cv2.cvtColor(output, cv2.COLOR_BGRA2RGBA)
        extension = 'png'
    else:
        display_img = cv2.cvtColor(output, cv2.COLOR_BGR2RGB)
        extension = 'jpg'

    out_filename = f"output_{rnd_string(8)}.{extension}"
    try:
        cv2.imwrite(out_filename, output)
        global last_file
        last_file = out_filename
    except Exception as e:
        print("Save error:", e)

    return display_img

#Add a batch upscaler that preserves filenames
def render_progress(pct: float, text: str = "") -> str:
    pct = max(0.0, min(100.0, float(pct)))
    bar = f"<div style='width:100%;border:1px solid #ddd;border-radius:6px;overflow:hidden;height:12px;'><div style='height:100%;width:{pct:.1f}%;background:#3b82f6;'></div></div>"
    label = f"<div style='font-size:12px;opacity:.8;margin-top:4px;'>{text} {pct:.1f}%</div>"
    return bar + label

@GPU()
def batch_realesrgan(
    files: list,             # from gr.Files (type='filepath')
    model_name: str,
    denoise_strength: float,
    face_enhance: bool,
    outscale: int,
    tile: int,
    batch_size: int = 16,
):
    """
    Processes multiple images in batches, preserves original file names for outputs,
    and returns (gallery, zip_file, details, progress_html) with streamed progress.
    """
    # Validate
    if not files or len(files) == 0:
        yield None, None, "No files uploaded.", render_progress(0, "Idle")
        return

    # Build upsampler once (much faster than per-image)
    upsampler, netscale, use_cuda, model_path = get_upsampler(model_name, outscale, tile=tile)

    # Optional: face enhancer (same as your single-image path)
    face_enhancer = None
    if face_enhance:
        from gfpgan import GFPGANer
        face_enhancer = GFPGANer(
            model_path='https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth',
            upscale=outscale,
            arch='clean',
            channel_multiplier=2,
            bg_upsampler=upsampler
        )

    # Prepare work/output dirs
    work = Path(tempfile.mkdtemp(prefix="batch_up_"))
    out_dir = work / "upscaled"
    out_dir.mkdir(parents=True, exist_ok=True)

    # Normalize list of input paths
    src_paths = [Path(f.name if hasattr(f, "name") else f) for f in files]

    total = len(src_paths)
    done = 0
    out_paths = []

    # If realesr-general-x4v3: support blending base + WDN via dni (optional)
    dni_weight = None
    if model_name == "realesr-general-x4v3":
        # Blend [base, WDN] with user's slider
        denoise_strength = float(denoise_strength)
        dni_weight = [1.0 - denoise_strength, denoise_strength]
        # RealESRGANer.enhance accepts dni_weight override via attribute on the instance
        try:
            upsampler.dni_weight = dni_weight
        except Exception:
            pass

    # Process in batches (I/O and PIL open are still per-file)
    for i in range(0, total, int(max(1, batch_size))):
        batch = src_paths[i:i + int(max(1, batch_size))]
        for src in batch:
            try:
                # Load as RGB consistently
                from PIL import Image
                with Image.open(src) as im:
                    img = im.convert("RGB")
                arr = numpy.array(img)
                arr = cv2.cvtColor(arr, cv2.COLOR_RGB2BGR)

                if face_enhancer:
                    _, _, output = face_enhancer.enhance(arr, has_aligned=False, only_center_face=False, paste_back=True)
                else:
                    output, _ = upsampler.enhance(arr, outscale=int(outscale))

                # Preserve original file name & (reasonable) extension
                orig_ext = src.suffix.lower()
                ext = orig_ext if orig_ext in (".png", ".jpg", ".jpeg") else ".png"
                out_path = out_dir / (src.stem + ext)

                # Save (keep alpha if produced, else RGB)
                if output.ndim == 3 and output.shape[2] == 4:
                    cv2.imwrite(str(out_path.with_suffix(".png")), output)  # 4ch → PNG
                    out_path = out_path.with_suffix(".png")
                else:
                    if ext in (".jpg", ".jpeg"):
                        cv2.imwrite(str(out_path), output, [int(cv2.IMWRITE_JPEG_QUALITY), 95])
                    else:
                        cv2.imwrite(str(out_path), output)  # PNG default
                out_paths.append(out_path)
            except Exception as e:
                # Continue on errors
                print(f"[batch] Error on {src}: {e}")
            finally:
                done += 1

        pct = (done / total) * 100.0 if total else 0.0
        remaining = max(0, total - done)
        msg = f"Upscaling… {done}/{total} done · {remaining} remaining (batch {(i//batch_size)+1}/{(total+batch_size-1)//batch_size})"
        yield None, None, msg, render_progress(pct, msg)

    if not out_paths:
        yield None, None, "No outputs produced.", render_progress(100, "Finished")
        return

    # Small even-sampled gallery for preview
    def _sample_even(seq, n=30):
        if not seq: return []
        if len(seq) <= n: return [str(p) for p in seq]
        step = (len(seq)-1) / (n-1)
        idxs = [round(i*step) for i in range(n)]
        seen, out = set(), []
        for i in idxs:
            if i not in seen:
                out.append(str(seq[int(i)])); seen.add(int(i))
        return out

    out_paths = sorted(out_paths)  # stable
    gallery = _sample_even(out_paths, 30)

    # Zip with same file names
    zip_path = work / "upscaled.zip"
    with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as zf:
        for p in out_paths:
            zf.write(p, arcname=p.name)

    details = f"Upscaled {len(out_paths)} images → {out_dir}"
    yield gallery, str(zip_path), details, render_progress(100, "Complete")


# ────────────────────────────────────────────────────────
# UI
# ────────────────────────────────────────────────────────
# ────────────────────────────────────────────────────────
# UI
# ────────────────────────────────────────────────────────
def main():
    with gr.Blocks(title="Real-ESRGAN Gradio Demo", theme="ParityError/Interstellar") as demo:
        gr.Markdown("## Image Upscaler")

        with gr.Accordion("Upscaling options", open=True):
            with gr.Row():
                model_name = gr.Dropdown(
                    label="Upscaler model",
                    choices=[
                        "RealESRGAN_x4plus",
                        "RealESRNet_x4plus",
                        "RealESRGAN_x4plus_anime_6B",
                        "RealESRGAN_x2plus",
                        "realesr-general-x4v3",
                    ],
                    value="RealESRGAN_x4plus",
                    show_label=True
                )
                denoise_strength = gr.Slider(
                    label="Denoise Strength (only for realesr-general-x4v3)",
                    minimum=0, maximum=1, step=0.1, value=0.5
                )
                outscale = gr.Slider(
                    label="Resolution upscale",
                    minimum=1, maximum=6, step=1, value=4, show_label=True
                )
                face_enhance = gr.Checkbox(label="Face Enhancement (GFPGAN)", value=False)

        model_tips = gr.Markdown(model_tip_text("RealESRGAN_x4plus"))

        with gr.Row():
            with gr.Group():
                input_image = gr.Image(label="Input Image", type="pil", image_mode="RGBA")
                input_image_properties = gr.Textbox(label="Image Properties", max_lines=1)
            output_image = gr.Image(label="Output Image", image_mode="RGBA")

        with gr.Row():
            reset_btn = gr.Button("Remove images")
            restore_btn = gr.Button("Upscale")

        input_image.change(fn=image_properties, inputs=input_image, outputs=input_image_properties)
        model_name.change(fn=model_tip_text, inputs=model_name, outputs=model_tips)

        restore_btn.click(
            fn=realesrgan,
            inputs=[input_image, model_name, denoise_strength, face_enhance, outscale],
            outputs=output_image
        )
        reset_btn.click(fn=reset, inputs=[], outputs=[output_image, input_image])

        # --- Batch Upscale (multi-image) ---
        gr.Markdown("### Batch Upscale")
        with gr.Accordion("Batch options", open=True):
            with gr.Row():
                batch_files = gr.Files(
                    label="Upload multiple images (PNG/JPG/JPEG)",
                    type="filepath",
                    file_types=[".png", ".jpg", ".jpeg"],
                    # file_count="multiple"  # optional: explicit in some versions
                )
            with gr.Row():
                batch_tile = gr.Number(label="Tile size (0/auto → 256)", value=256, precision=0)
                batch_size = gr.Number(label="Batch size (images per batch)", value=16, precision=0)

        with gr.Row():
            batch_btn = gr.Button("Upscale Batch", variant="primary")

        batch_prog = gr.HTML(render_progress(0.0, "Idle"))
        batch_gallery = gr.Gallery(label="Preview (sampled 30)", columns=6, height=420)
        batch_zip = gr.File(label="Download upscaled.zip")
        batch_details = gr.Markdown("")

        # Wire it up (generator → streaming)
        batch_btn.click(
            fn=batch_realesrgan,
            inputs=[batch_files, model_name, denoise_strength, face_enhance, outscale, batch_tile, batch_size],
            outputs=[batch_gallery, batch_zip, batch_details, batch_prog],
        )

        gr.Markdown("")  # spacer

    return demo


if __name__ == "__main__":
    demo = main()
    demo.queue().launch(ssr_mode=False)  # set share=True if you want a public link