app.py

import torch
from torch import nn
import torch.nn.functional as F
import torchvision.transforms.functional as TF

from torch import Tensor
import spaces

import numpy as np
from PIL import Image
import gradio as gr
from matplotlib import cm
from huggingface_hub import hf_hub_download
from warnings import warn

from models import get_model


mean = (0.485, 0.456, 0.406)
std = (0.229, 0.224, 0.225)
alpha = 0.8
EPS = 1e-8
loaded_model = None
current_model_config = {"variant": None, "dataset": None, "metric": None}

pretrained_models = [
    "ZIP-B @ ShanghaiTech A @ MAE", "ZIP-B @ ShanghaiTech A @ NAE",
    "ZIP-B @ ShanghaiTech B @ MAE", "ZIP-B @ ShanghaiTech B @ NAE",
    "ZIP-B @ UCF-QNRF @ MAE", "ZIP-B @ UCF-QNRF @ NAE",
    "ZIP-B @ NWPU-Crowd @ MAE", "ZIP-B @ NWPU-Crowd @ NAE",
    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
    "ZIP-S @ ShanghaiTech A @ MAE", "ZIP-S @ ShanghaiTech A @ NAE",
    "ZIP-S @ ShanghaiTech B @ MAE", "ZIP-S @ ShanghaiTech B @ NAE",
    "ZIP-S @ UCF-QNRF @ MAE", "ZIP-S @ UCF-QNRF @ NAE",
    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
    "ZIP-T @ ShanghaiTech A @ MAE", "ZIP-T @ ShanghaiTech A @ NAE",
    "ZIP-T @ ShanghaiTech B @ MAE", "ZIP-T @ ShanghaiTech B @ NAE",
    "ZIP-T @ UCF-QNRF @ MAE", "ZIP-T @ UCF-QNRF @ NAE",
    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
    "ZIP-N @ ShanghaiTech A @ MAE", "ZIP-N @ ShanghaiTech A @ NAE",
    "ZIP-N @ ShanghaiTech B @ MAE", "ZIP-N @ ShanghaiTech B @ NAE",
    "ZIP-N @ UCF-QNRF @ MAE", "ZIP-N @ UCF-QNRF @ NAE",
    "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━",
    "ZIP-P @ ShanghaiTech A @ MAE", "ZIP-P @ ShanghaiTech A @ NAE",
    "ZIP-P @ ShanghaiTech B @ MAE", "ZIP-P @ ShanghaiTech B @ NAE",
    "ZIP-P @ UCF-QNRF @ MAE", "ZIP-P @ UCF-QNRF @ NAE",
]

# -----------------------------
# Model management functions
# -----------------------------
def update_model_if_needed(variant_dataset_metric: str):
    """
    Load a new model only if the configuration has changed.
    """
    global loaded_model, current_model_config
    
    # 如果是分割线，则跳过
    if "━━━━━━" in variant_dataset_metric:
        return "Please select a valid model configuration"
    
    parts = variant_dataset_metric.split(" @ ")
    if len(parts) != 3:
        return "Invalid model configuration format"
        
    variant, dataset, metric = parts[0], parts[1], parts[2].lower()

    if dataset == "ShanghaiTech A":
        dataset_name = "sha"
    elif dataset == "ShanghaiTech B":
        dataset_name = "shb"
    elif dataset == "UCF-QNRF":
        dataset_name = "qnrf"
    elif dataset == "NWPU-Crowd":
        dataset_name = "nwpu"
    else:
        return f"Unknown dataset: {dataset}"

    # 只更新配置，不在主进程中加载模型
    if (current_model_config["variant"] != variant or 
        current_model_config["dataset"] != dataset_name or 
        current_model_config["metric"] != metric):
        
        print(f"Model configuration updated: {variant} @ {dataset} with {metric} metric")
        current_model_config = {"variant": variant, "dataset": dataset_name, "metric": metric}
        loaded_model = None  # 重置模型，将在GPU进程中重新加载
        return f"Model configuration set: {variant} @ {dataset} ({metric})"
    else:
        print(f"Model configuration unchanged: {variant} @ {dataset} with {metric} metric")
        return f"Model configuration: {variant} @ {dataset} ({metric})"


# -----------------------------
# Define the model architecture
# -----------------------------
def load_model(variant: str, dataset: str = "ShanghaiTech B", metric: str = "mae"):
    """ Load the model weights from the Hugging Face Hub."""
    # global loaded_model
    # Build model

    model_info_path = hf_hub_download(
        repo_id=f"Yiming-M/{variant}",
        filename=f"checkpoints/{dataset}/best_{metric}.pth",
    )

    model = get_model(model_info_path=model_info_path)
    model.eval()
    # loaded_model = model
    return model


def _calc_size(
    img_w: int,
    img_h: int,
    min_size: int,
    max_size: int,
    base: int = 32
):
    """
    This function generates a new size for an image while keeping the aspect ratio. The new size should be within the given range (min_size, max_size).

    Args:
        img_w (int): The width of the image.
        img_h (int): The height of the image.
        min_size (int): The minimum size of the edges of the image.
        max_size (int): The maximum size of the edges of the image.
        # base (int): The base number to which the new size should be a multiple of.
    """
    assert min_size % base == 0, f"min_size ({min_size}) must be a multiple of {base}"
    if max_size != float("inf"):
        assert max_size % base == 0, f"max_size ({max_size}) must be a multiple of {base} if provided"

    assert min_size <= max_size, f"min_size ({min_size}) must be less than or equal to max_size ({max_size})"

    aspect_ratios = (img_w / img_h, img_h / img_w)
    if min_size / max_size <= min(aspect_ratios) <= max(aspect_ratios) <= max_size / min_size:  # possible to resize and preserve the aspect ratio
        if min_size <= min(img_w, img_h) <= max(img_w, img_h) <= max_size:  # already within the range, no need to resize
            ratio = 1.
        elif min(img_w, img_h) < min_size:  # smaller than the minimum size, resize to the minimum size
            ratio = min_size / min(img_w, img_h)
        else:  # larger than the maximum size, resize to the maximum size
            ratio = max_size / max(img_w, img_h)

        new_w, new_h = int(round(img_w * ratio / base) * base), int(round(img_h * ratio / base) * base)
        new_w = max(min_size, min(max_size, new_w))
        new_h = max(min_size, min(max_size, new_h))
        return new_w, new_h

    else:  # impossible to resize and preserve the aspect ratio
        msg = f"Impossible to resize {img_w}x{img_h} image while preserving the aspect ratio to a size within the range ({min_size}, {max_size}). Will not limit the maximum size."
        warn(msg)
        return _calc_size(img_w, img_h, min_size, float("inf"), base)
    

# -----------------------------
# Preprocessing function
# -----------------------------
# Adjust the image transforms to match what your model expects.
def transform(image: Image.Image, dataset_name: str) -> Tensor:
    assert isinstance(image, Image.Image), "Input must be a PIL Image"
    image_tensor = TF.to_tensor(image)

    if dataset_name == "sha":
        min_size = 448
        max_size = float("inf")
    elif dataset_name == "shb":
        min_size = 448
        max_size = float("inf")
    elif dataset_name == "qnrf":
        min_size = 448
        max_size = 2048
    elif dataset_name == "nwpu":
        min_size = 448
        max_size = 3072

    image_height, image_width = image_tensor.shape[-2:]
    new_width, new_height = _calc_size(
        img_w=image_width,
        img_h=image_height,
        min_size=min_size,
        max_size=max_size,
        base=32
    )
    if new_height != image_height or new_width != image_width:
        image_tensor = TF.resize(image_tensor, size=(new_height, new_width), interpolation=TF.InterpolationMode.BICUBIC, antialias=True)

    image_tensor = TF.normalize(image_tensor, mean=mean, std=std)
    return image_tensor.unsqueeze(0)  # Add batch dimension


def _sliding_window_predict(
    model: nn.Module,
    image: Tensor,
    window_size: int,
    stride: int, 
    max_num_windows: int = 256
):
    assert len(image.shape) == 4, f"Image must be a 4D tensor (1, c, h, w), got {image.shape}"
    window_size = (int(window_size), int(window_size)) if isinstance(window_size, (int, float)) else window_size
    stride = (int(stride), int(stride)) if isinstance(stride, (int, float)) else stride
    window_size = tuple(window_size)
    stride = tuple(stride)
    assert isinstance(window_size, tuple) and len(window_size) == 2 and window_size[0] > 0 and window_size[1] > 0, f"Window size must be a positive integer tuple (h, w), got {window_size}"
    assert isinstance(stride, tuple) and len(stride) == 2 and stride[0] > 0 and stride[1] > 0, f"Stride must be a positive integer tuple (h, w), got {stride}"
    assert stride[0] <= window_size[0] and stride[1] <= window_size[1], f"Stride must be smaller than window size, got {stride} and {window_size}"

    image_height, image_width = image.shape[-2:]
    window_height, window_width = window_size
    assert image_height >= window_height and image_width >= window_width, f"Image size must be larger than window size, got image size {image.shape} and window size {window_size}"
    stride_height, stride_width = stride

    num_rows = int(np.ceil((image_height - window_height) / stride_height) + 1)
    num_cols = int(np.ceil((image_width - window_width) / stride_width) + 1)

    if hasattr(model, "block_size"):
        block_size = model.block_size
    elif hasattr(model, "module") and hasattr(model.module, "block_size"):
        block_size = model.module.block_size
    else:
        raise ValueError("Model must have block_size attribute")
    assert window_height % block_size == 0 and window_width % block_size == 0, f"Window size must be divisible by block size, got {window_size} and {block_size}"

    windows = []
    for i in range(num_rows):
        for j in range(num_cols):
            x_start, y_start = i * stride_height, j * stride_width
            x_end, y_end = x_start + window_height, y_start + window_width
            if x_end > image_height:
                x_start, x_end = image_height - window_height, image_height
            if y_end > image_width:
                y_start, y_end = image_width - window_width, image_width

            window = image[:, :, x_start:x_end, y_start:y_end]
            windows.append(window)

    windows = torch.cat(windows, dim=0).to(image.device)  # batched windows, shape: (num_windows, c, h, w)
    
    model.eval()
    pi_maps, lambda_maps = [], []
    for i in range(0, len(windows), max_num_windows):
        with torch.no_grad():
            image_feats = model.backbone(windows[i: min(i + max_num_windows, len(windows))])
            pi_image_feats, lambda_image_feats = model.pi_head(image_feats), model.lambda_head(image_feats)
            pi_image_feats = F.normalize(pi_image_feats.permute(0, 2, 3, 1), p=2, dim=-1)  # shape (B, H, W, C)
            lambda_image_feats = F.normalize(lambda_image_feats.permute(0, 2, 3, 1), p=2, dim=-1)  # shape (B, H, W, C)

            pi_text_feats, lambda_text_feats = model.pi_text_feats, model.lambda_text_feats
            pi_logit_scale, lambda_logit_scale = model.pi_logit_scale.exp(), model.lambda_logit_scale.exp()

            pi_logit_map = pi_logit_scale * pi_image_feats @ pi_text_feats.t()  # (B, H, W, 2), logits per image
            lambda_logit_map = lambda_logit_scale * lambda_image_feats @ lambda_text_feats.t()  # (B, H, W, N - 1), logits per image

            pi_logit_map =  pi_logit_map.permute(0, 3, 1, 2)  # (B, 2, H, W)
            lambda_logit_map = lambda_logit_map.permute(0, 3, 1, 2)  # (B, N - 1, H, W)

            lambda_map = (lambda_logit_map.softmax(dim=1) * model.bin_centers[:, 1:]).sum(dim=1, keepdim=True)  # (B, 1, H, W)
            pi_map = pi_logit_map.softmax(dim=1)[:, 0:1]  # (B, 1, H, W)

            pi_maps.append(pi_map.cpu().numpy())
            lambda_maps.append(lambda_map.cpu().numpy())

    # assemble the density map
    pi_maps = np.concatenate(pi_maps, axis=0)  # shape: (num_windows, 1, H, W)
    lambda_maps = np.concatenate(lambda_maps, axis=0)  # shape: (num_windows, 1, H, W)
    assert pi_maps.shape == lambda_maps.shape, f"pi_maps and lambda_maps must have the same shape, got {pi_maps.shape} and {lambda_maps.shape}"

    pi_map = np.zeros((pi_maps.shape[1], image_height // block_size, image_width // block_size), dtype=np.float32)
    lambda_map = np.zeros((lambda_maps.shape[1], image_height // block_size, image_width // block_size), dtype=np.float32)
    count_map = np.zeros((pi_maps.shape[1], image_height // block_size, image_width // block_size), dtype=np.float32)
    idx = 0
    for i in range(num_rows):
        for j in range(num_cols):
            x_start, y_start = i * stride_height, j * stride_width
            x_end, y_end = x_start + window_height, y_start + window_width
            if x_end > image_height:
                x_start, x_end = image_height - window_height, image_height
            if y_end > image_width:
                y_start, y_end = image_width - window_width, image_width

            pi_map[:, (x_start // block_size): (x_end // block_size), (y_start // block_size): (y_end // block_size)] += pi_maps[idx, :, :, :]
            lambda_map[:, (x_start // block_size): (x_end // block_size), (y_start // block_size): (y_end // block_size)] += lambda_maps[idx, :, :, :]
            count_map[:, (x_start // block_size): (x_end // block_size), (y_start // block_size): (y_end // block_size)] += 1.
            idx += 1

    # average the density map
    pi_map /= count_map
    lambda_map /= count_map
    
    # convert to Tensor and reshape
    pi_map = torch.from_numpy(pi_map).unsqueeze(0)  # shape: (1, 1, H // block_size, W // block_size)
    lambda_map = torch.from_numpy(lambda_map).unsqueeze(0)  # shape: (1, 1, H // block_size, W // block_size)
    return pi_map, lambda_map


# -----------------------------
# Inference function
# -----------------------------
@spaces.GPU(duration=120)
def predict(image: Image.Image, variant_dataset_metric: str):
    """
    Given an input image, preprocess it, run the model to obtain a density map,
    compute the total crowd count, and prepare the density map for display.
    """
    global loaded_model, current_model_config
    
    # 在GPU进程中定义device
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    
    # 如果选择的是分割线，返回错误信息
    if "━━━━━━" in variant_dataset_metric:
        return image, None, None, "⚠️ Please select a valid model configuration", None, None, None
    
    parts = variant_dataset_metric.split(" @ ")
    if len(parts) != 3:
        return image, None, None, "❌ Invalid model configuration format", None, None, None
        
    variant, dataset, metric = parts[0], parts[1], parts[2].lower()

    if dataset == "ShanghaiTech A":
        dataset_name = "sha"
    elif dataset == "ShanghaiTech B":
        dataset_name = "shb"
    elif dataset == "UCF-QNRF":
        dataset_name = "qnrf"
    elif dataset == "NWPU-Crowd":
        dataset_name = "nwpu"
    else:
        return image, None, None, f"❌ Unknown dataset: {dataset}", None, None, None

    # 在GPU进程中加载模型（如果需要）
    if (loaded_model is None or 
        current_model_config["variant"] != variant or 
        current_model_config["dataset"] != dataset_name or 
        current_model_config["metric"] != metric):
        
        print(f"Loading model in GPU process: {variant} @ {dataset} with {metric} metric")
        loaded_model = load_model(variant=variant, dataset=dataset_name, metric=metric)
        current_model_config = {"variant": variant, "dataset": dataset_name, "metric": metric}

    if not hasattr(loaded_model, "input_size"):
        if dataset_name == "sha":
            loaded_model.input_size = 224
        elif dataset_name == "shb":
            loaded_model.input_size = 448
        elif dataset_name == "qnrf":
            loaded_model.input_size = 672
        elif dataset_name == "nwpu":
            loaded_model.input_size = 672
    elif isinstance(loaded_model.input_size, (list, tuple)):
        loaded_model.input_size = loaded_model.input_size[0]  # Use the first element if it's a list or tuple
    else:
        assert isinstance(loaded_model.input_size, (int, float)), f"input_size must be an int or float, got {type(loaded_model.input_size)}"

    loaded_model.to(device)

    # Preprocess the image
    input_width, input_height = image.size
    image_tensor = transform(image, dataset_name).to(device)  # shape: (1, 3, H, W)

    input_size = loaded_model.input_size
    image_height, image_width = image_tensor.shape[-2:]
    aspect_ratio = image_width / image_height
    if image_height < input_size:
        new_height = input_size
        new_width = int(new_height * aspect_ratio)
        image_tensor = F.interpolate(image_tensor, size=(new_height, new_width), mode="bicubic", align_corners=False, antialias=True)
        image_height, image_width = new_height, new_width
    if image_width < input_size:
        new_width = input_size
        new_height = int(new_width / aspect_ratio)
        image_tensor = F.interpolate(image_tensor, size=(new_height, new_width), mode="bicubic", align_corners=False, antialias=True)
        image_height, image_width = new_height, new_width
    
    with torch.no_grad():
        if hasattr(loaded_model, "num_vpt") and loaded_model.num_vpt is not None and loaded_model.num_vpt > 0:  # For ViT models, use sliding window prediction
            # For ViT models with VPT
            pi_map, lambda_map = _sliding_window_predict(
                model=loaded_model,
                image=image_tensor,
                window_size=input_size,
                stride=input_size
            )
        
        elif hasattr(loaded_model, "pi_text_feats") and hasattr(loaded_model, "lambda_text_feats") and loaded_model.pi_text_feats is not None and loaded_model.lambda_text_feats is not None:  # For other CLIP-based models
            image_feats = loaded_model.backbone(image_tensor)
            # image_feats = F.normalize(image_feats.permute(0, 2, 3, 1), p=2, dim=-1)  # shape (B, H, W, C)
            pi_image_feats, lambda_image_feats = loaded_model.pi_head(image_feats), loaded_model.lambda_head(image_feats)
            pi_image_feats = F.normalize(pi_image_feats.permute(0, 2, 3, 1), p=2, dim=-1)  # shape (B, H, W, C)
            lambda_image_feats = F.normalize(lambda_image_feats.permute(0, 2, 3, 1), p=2, dim=-1)  # shape (B, H, W, C)

            pi_text_feats, lambda_text_feats = loaded_model.pi_text_feats, loaded_model.lambda_text_feats
            pi_logit_scale, lambda_logit_scale = loaded_model.pi_logit_scale.exp(), loaded_model.lambda_logit_scale.exp()

            pi_logit_map = pi_logit_scale * pi_image_feats @ pi_text_feats.t()  # (B, H, W, 2), logits per image
            lambda_logit_map = lambda_logit_scale * lambda_image_feats @ lambda_text_feats.t()  # (B, H, W, N - 1), logits per image

            pi_logit_map =  pi_logit_map.permute(0, 3, 1, 2)  # (B, 2, H, W)
            lambda_logit_map = lambda_logit_map.permute(0, 3, 1, 2)  # (B, N - 1, H, W)

            lambda_map = (lambda_logit_map.softmax(dim=1) * loaded_model.bin_centers[:, 1:]).sum(dim=1, keepdim=True)  # (B, 1, H, W)
            pi_map = pi_logit_map.softmax(dim=1)[:, 0:1]  # (B, 1, H, W)
        
        else: # For non-CLIP models
            x = loaded_model.backbone(image_tensor)
            logit_pi_map = loaded_model.pi_head(x)  # shape: (B, 2, H, W)
            logit_map = loaded_model.bin_head(x)  # shape: (B, C, H, W)
            lambda_map= (logit_map.softmax(dim=1) * loaded_model.bin_centers[:, 1:]).sum(dim=1, keepdim=True)  # shape: (B, 1, H, W)
            pi_map = logit_pi_map.softmax(dim=1)[:, 0:1]  # shape: (B, 1, H, W)

        
        den_map = (1.0 - pi_map) * lambda_map  # shape: (B, 1, H, W)
        count = den_map.sum().item()

        strucrual_zero_map = F.interpolate(
            pi_map, size=(input_height, input_width), mode="bilinear", align_corners=False, antialias=True
        ).cpu().squeeze().numpy()

        lambda_map = F.interpolate(
            lambda_map, size=(input_height, input_width), mode="bilinear", align_corners=False, antialias=True
        ).cpu().squeeze().numpy()

        den_map = F.interpolate(
            den_map, size=(input_height, input_width), mode="bilinear", align_corners=False, antialias=True
        ).cpu().squeeze().numpy()
    
    sampling_zero_map = (1.0 - strucrual_zero_map) * np.exp(-lambda_map)
    complete_zero_map = strucrual_zero_map + sampling_zero_map

    # Normalize maps for display purposes
    def normalize_map(x: np.ndarray) -> np.ndarray:
        """ Normalize the map to [0, 1] range for visualization. """
        x_min = np.min(x)
        x_max = np.max(x)
        if x_max - x_min < EPS:
            return np.zeros_like(x)
        return (x - x_min) / (x_max - x_min + EPS)
    
    # strucrual_zero_map = normalize_map(strucrual_zero_map)
    # sampling_zero_map = normalize_map(sampling_zero_map)
    # lambda_map = normalize_map(lambda_map)
    # den_map = normalize_map(den_map)
    # complete_zero_map = normalize_map(complete_zero_map)
    
    # Apply a colormap for better visualization
    # Options: 'viridis', 'plasma', 'hot', 'inferno', 'jet' (recommended)
    colormap = cm.get_cmap("jet")

    # The colormap returns values in [0,1]. Scale to [0,255] and convert to uint8.
    den_map = (colormap(den_map) * 255).astype(np.uint8)
    strucrual_zero_map = (colormap(strucrual_zero_map) * 255).astype(np.uint8)
    sampling_zero_map = (colormap(sampling_zero_map) * 255).astype(np.uint8)
    lambda_map = (colormap(lambda_map) * 255).astype(np.uint8)
    complete_zero_map = (colormap(complete_zero_map) * 255).astype(np.uint8)

    # Convert to PIL images
    den_map = Image.fromarray(den_map).convert("RGBA")
    strucrual_zero_map = Image.fromarray(strucrual_zero_map).convert("RGBA")
    sampling_zero_map = Image.fromarray(sampling_zero_map).convert("RGBA")
    lambda_map = Image.fromarray(lambda_map).convert("RGBA")
    complete_zero_map = Image.fromarray(complete_zero_map).convert("RGBA")
    
    # Ensure the original image is in RGBA format.
    image_rgba = image.convert("RGBA")

    den_map = Image.blend(image_rgba, den_map, alpha=alpha)
    strucrual_zero_map = Image.blend(image_rgba, strucrual_zero_map, alpha=alpha)
    sampling_zero_map = Image.blend(image_rgba, sampling_zero_map, alpha=alpha)
    lambda_map = Image.blend(image_rgba, lambda_map, alpha=alpha)
    complete_zero_map = Image.blend(image_rgba, complete_zero_map, alpha=alpha)
    
    # 格式化计数显示
    count_display = f"👥 {round(count, 2)} people detected"
    if count < 1:
        count_display = "👤 Less than 1 person detected"
    elif count == 1:
        count_display = "👤 1 person detected"
    elif count < 10:
        count_display = f"👥 {round(count, 1)} people detected"
    else:
        count_display = f"👥 {round(count)} people detected"
    
    return image, den_map, lambda_map, count_display, strucrual_zero_map, sampling_zero_map, complete_zero_map


# -----------------------------
# Build Gradio Interface using Blocks for a two-column layout
# -----------------------------
css = """
/* 基础样式 - 保持功能性 */
.gradio-container {
    max-width: 1600px;
    margin: 0 auto;
    padding: 20px;
}

/* 简单的分割线样式 */
option[value*="━━━━━━"] {
    color: #999;
    background-color: #f0f0f0;
    text-align: center;
}

/* 基础组件样式 */
.gr-group {
    background: white;
    border-radius: 8px;
    padding: 16px;
    margin: 8px 0;
    box-shadow: 0 2px 4px rgba(0,0,0,0.1);
}

.gr-button {
    background: #3b82f6;
    color: white;
    border: none;
    border-radius: 6px;
    padding: 12px 24px;
    font-weight: 600;
}

.gr-image {
    border-radius: 8px;
    height: 400px;
}

/* 响应式设计 */
@media (max-width: 768px) {
    .gradio-container {
        padding: 12px;
    }
    
    .gr-image {
        height: 300px;
    }
}
"""

with gr.Blocks(css=css, theme=gr.themes.Soft(), title="ZIP Crowd Counting") as demo:
    gr.Markdown("""
    # 🎯 Crowd Counting by ZIP
    ### Upload an image and get precise crowd density predictions with ZIP models!
    """)
    
    # 添加信息面板
    with gr.Accordion("ℹ️ About ZIP", open=False):
        gr.Markdown("""
        **ZIP (Zero-Inflated Poisson)** is a framework designed for crowd counting, a task where the goal is to estimate how many people are present in an image. It was introduced in the paper [ZIP: Scalable Crowd Counting via Zero-Inflated Poisson Modeling](https://arxiv.org/abs/2506.19955).
        ZIP is based on a simple idea: not all empty areas in an image mean the same thing. Some regions are empty because there are truly no people there (like walls or sky), while others are places where people could appear but just happen not to in this particular image. ZIP separates these two cases using two prediction heads:
        - **Structural Zeros**: These are regions that naturally never contain people (e.g., the background or torso areas). These are handled by the π head.
        - **Sampling Zeros**: These are regions where people could appear but don't in this image. These are modeled by the λ head.
        
        By separating *where* people are likely to be from *how many* are present, ZIP produces more accurate and interpretable crowd estimates, especially in scenes with large empty spaces or varied crowd densities.
        
        Choose from different model variants: **ZIP-B** (Base), **ZIP-S** (Small), **ZIP-T** (Tiny), **ZIP-N** (Nano), **ZIP-P** (Pico)
        """)

    # 第二行：模型配置、状态和预测结果（三列等宽）
    with gr.Row():
        with gr.Column(scale=1):
            with gr.Group():
                model_dropdown = gr.Dropdown(
                    choices=pretrained_models,
                    value="ZIP-B @ NWPU-Crowd @ MAE",
                    label="🎛️ Select Model & Dataset",
                    info="Choose model variant, dataset, and evaluation metric"
                )
        
        with gr.Column(scale=1):
            with gr.Group():
                model_status = gr.Textbox(
                    label="📊 Model Status",
                    value="🔄 No model loaded",
                    interactive=False,
                    elem_classes=["status-display"],
                    lines=3
                )
        
        with gr.Column(scale=1):
            with gr.Group():
                output_text = gr.Textbox(
                    label="🧙 Predicted Count",
                    value="",
                    interactive=False,
                    info="Total number of people detected",
                    lines=3
                )

    # 第三行：主要图像（输入图像、密度图、Lambda图）
    with gr.Row():
        with gr.Column(scale=1):
            with gr.Group():
                input_img = gr.Image(
                    label="📸 Upload Image", 
                    sources=["upload", "clipboard"], 
                    type="pil",
                    height=400
                )
                submit_btn = gr.Button(
                    "🚀 Analyze Crowd", 
                    variant="primary",
                    size="lg"
                )
        
        with gr.Column(scale=1):
            with gr.Group():
                output_den_map = gr.Image(
                    label="🎯 Predicted Density Map", 
                    type="pil", 
                    height=400
                )
        
        with gr.Column(scale=1):
            with gr.Group():
                output_lambda_map = gr.Image(
                    label="📈 Lambda Map", 
                    type="pil", 
                    height=400
                )
    
    # 第四行：Zero Analysis - 全宽，内部三列等宽
    with gr.Group():
        gr.Markdown("### 🔍 Zero Analysis")
        gr.Markdown("*Explore different types of zero predictions in crowd analysis*")
        with gr.Row():
            with gr.Column(scale=1):
                output_structural_zero_map = gr.Image(
                    label="🏗️ Structural Zero Map", 
                    type="pil", 
                    height=400,
                    elem_classes=["zero-analysis-image"]
                )
            
            with gr.Column(scale=1):
                output_sampling_zero_map = gr.Image(
                    label="📊 Sampling Zero Map", 
                    type="pil", 
                    height=400,
                    elem_classes=["zero-analysis-image"]
                )
            
            with gr.Column(scale=1):
                output_complete_zero_map = gr.Image(
                    label="👺 Complete Zero Map", 
                    type="pil", 
                    height=400,
                    elem_classes=["zero-analysis-image"]
                )

    # 当模型变化时，自动更新模型
    def on_model_change(variant_dataset_metric):
        # 如果选择的是分割线，保持当前选择不变
        if "━━━━━━" in variant_dataset_metric:
            return "⚠️ Please select a valid model configuration"
        result = update_model_if_needed(variant_dataset_metric)
        if "Model loaded:" in result:
            return f"✅ {result}"
        elif "Model already loaded:" in result:
            return f"🔄 {result}"
        else:
            return f"❌ {result}"
    
    model_dropdown.change(
        fn=on_model_change,
        inputs=[model_dropdown],
        outputs=[model_status]
    )

    # 页面加载时设置默认模型配置（不在主进程中加载模型）
    demo.load(
        fn=lambda: f"✅ {update_model_if_needed('ZIP-B @ NWPU-Crowd @ MAE')}",
        outputs=[model_status]
    )

    submit_btn.click(
        fn=predict,
        inputs=[input_img, model_dropdown],
        outputs=[input_img, output_den_map, output_lambda_map, output_text, output_structural_zero_map, output_sampling_zero_map, output_complete_zero_map]
    )

    # 美化示例区域
    with gr.Accordion("🖼️ Try Example Images", open=True):
        gr.Markdown("**Click on any example below to test the model:**")
        gr.Examples(
            examples=[
                ["example1.jpg"], ["example2.jpg"], ["example3.jpg"], ["example4.jpg"],
                ["example5.jpg"], ["example6.jpg"], ["example7.jpg"], ["example8.jpg"],
                ["example9.jpg"], ["example10.jpg"], ["example11.jpg"], ["example12.jpg"]
            ],
            inputs=input_img,
            label="📚 Example Gallery",
            examples_per_page=12
        )
    
    # 添加使用说明
    with gr.Accordion("📖 How to Use", open=False):
        gr.Markdown("""
        ### Step-by-step Guide:
        
        1. **🎛️ Select Model**: Choose your preferred model variant, pre-trained dataset, and evaluation metric from the dropdown
        2. **📸 Upload Image**: Click the image area to upload your crowd photo or use clipboard
        3. **🚀 Analyze**: Click the "Analyze Crowd" button to start processing
        4. **📊 View Results**: Examine the density maps and crowd count in the output panels
        
        ### Understanding the Outputs:
        
        **📊 Main Results:**
        - **🎯 Density Map**: Shows where people are located with color intensity, modeled by (1-π) * λ
        - **🧙 Predicted Count**: Total number of people detected in the image
        
        **🔍 Zero Analysis:**
        - **🏗️ Structural Zero Map**: Indicates regions that structurally cannot contain head annotations (e.g., walls, sky, torso, or background). These are governed by the π head, which estimates the probability that a region never contains people.
        - **📊 Sampling Zero Map**: Shows areas where people could be present but happen not to appear in the current image. These zeros are modeled by (1-π) * exp(-λ), where the expected count λ is near zero.
        - **👺 Complete Zero Map**: A combined visualization of zero probabilities, capturing both structural and sampling zeros. This map reflects overall non-crowd likelihood per region.
        
        **🔥 Hotspots:**
        - **📈 Lambda Map**: Highlights areas with high expected crowd density. Each value represents the expected number of people in that region, modeled by the Poisson intensity (λ). This map focuses on *how many* people are likely to be present, **WITHOUT** assuming people could appear there. ⚠️ Lambda Map **NEEDS** to be combined with Structural Zero Map by (1-π) * λ to produce the final density map.
        """)
    
    # 添加技术信息
    with gr.Accordion("🔬 Technical Details", open=False):
        gr.Markdown("""
        ### Model Variants:
        - **ZIP-B**: Base model with best performance
        - **ZIP-S**: Small model for faster inference
        - **ZIP-T**: Tiny model for resource-constrained environments
        - **ZIP-N**: Nano model for mobile applications
        - **ZIP-P**: Pico model for edge devices
        
        ### Datasets:
        - **ShanghaiTech A**: Dense, low-resolution crowd scenes
        - **ShanghaiTech B**: Sparse, high-resolution crowd scenes
        - **UCF-QNRF**: Dense, ultra high-resolution crowd images
        - **NWPU-Crowd**: Largest ultra high-resolution crowd counting dataset
        
        ### Metrics:
        - **MAE**: Mean Absolute Error - average counting error
        - **NAE**: Normalized Absolute Error - relative counting error
        """)

demo.launch(
    server_name="0.0.0.0",
    server_port=7860,
    show_api=False,
    share=False
)