pokouqiege/models/line_detect/heads/head_losses.py

import torch
from matplotlib import pyplot as plt

import torch.nn.functional as F
from torch import nn
from torch.cuda import device


class DiceLoss(nn.Module):
    def __init__(self, smooth=1.):
        super(DiceLoss, self).__init__()
        self.smooth = smooth

    def forward(self, logits, targets):
        probs = torch.sigmoid(logits)
        probs = probs.view(-1)
        targets = targets.view(-1).float()

        intersection = (probs * targets).sum()
        dice = (2. * intersection + self.smooth) / (probs.sum() + targets.sum() + self.smooth)
        return 1. - dice
bce_loss = nn.BCEWithLogitsLoss()
dice_loss = DiceLoss()


def combined_loss(preds, targets, alpha=0.5):
    bce = bce_loss(preds, targets)
    d = dice_loss(preds, targets)
    return alpha * bce + (1 - alpha) * d

def features_align(features, proposals, img_size):
    print(f'features_align features:{features.shape},proposals:{len(proposals)}')

    align_feat_list = []

    for feat, proposals_per_img in zip(features, proposals):
        print(f'feature_align feat:{feat.shape}, proposals_per_img:{proposals_per_img.shape}')
        if proposals_per_img.shape[0]>0:
            feat = feat.unsqueeze(0)
            for proposal in proposals_per_img:
                align_feat = torch.zeros_like(feat)
                # print(f'align_feat:{align_feat.shape}')
                x1, y1, x2, y2 = map(lambda v: int(v.item()), proposal)
                # 将每个proposal框内的部分赋值到align_feats对应位置
                align_feat[:, :, y1:y2 + 1, x1:x2 + 1] = feat[:, :, y1:y2 + 1, x1:x2 + 1]
                align_feat_list.append(align_feat)

    # print(f'align_feat_list:{align_feat_list}')
    if len(align_feat_list) > 0:
        feats_tensor = torch.cat(align_feat_list)

        print(f'align features :{feats_tensor.shape}')
    else:
        feats_tensor = None

    return feats_tensor
def normalize_tensor(t):
    return (t - t.min()) / (t.max() - t.min() + 1e-6)

def line_length(lines):
    """
    计算每条线段的长度
    lines: [N, 2, 2] 表示 N 条线段，每条线段由两个点组成
    返回: [N]
    """
    return torch.norm(lines[:, 1] - lines[:, 0], dim=-1)

def line_direction(lines):
    """
    计算每条线段的单位方向向量
    lines: [N, 2, 2]
    返回: [N, 2] 单位方向向量
    """
    vec = lines[:, 1] - lines[:, 0]
    return F.normalize(vec, dim=-1)

def angle_loss_cosine(pred_dir, gt_dir):
    """
    使用 cosine similarity 计算方向差异
    pred_dir: [N, 2]
    gt_dir: [N, 2]
    返回: [N]
    """
    cos_sim = torch.sum(pred_dir * gt_dir, dim=-1).clamp(-1.0, 1.0)
    return 1.0 - cos_sim  # 或者 torch.acos(cos_sim) / pi 也可


def line_length(lines):
        """
        计算每条线段的长度
        lines: [N, 2, 2] 表示 N 条线段，每条线段由两个点组成
        返回: [N]
        """
        return torch.norm(lines[:, 1] - lines[:, 0], dim=-1)

def line_direction(lines):
        """
        计算每条线段的单位方向向量
        lines: [N, 2, 2]
        返回: [N, 2] 单位方向向量
        """
        vec = lines[:, 1] - lines[:, 0]
        return F.normalize(vec, dim=-1)

def angle_loss_cosine(pred_dir, gt_dir):
        """
        使用 cosine similarity 计算方向差异
        pred_dir: [N, 2]
        gt_dir: [N, 2]
        返回: [N]
        """
        cos_sim = torch.sum(pred_dir * gt_dir, dim=-1).clamp(-1.0, 1.0)
        return 1.0 - cos_sim  # 或者 torch.acos(cos_sim) / pi 也可


def single_point_to_heatmap(keypoints, rois, heatmap_size):
    # type: (Tensor, Tensor, int) -> Tensor
    print(f'rois:{rois.shape}')
    print(f'heatmap_size:{heatmap_size}')


    print(f'keypoints.shape:{keypoints.shape}')
    # batch_size, num_keypoints, _ = keypoints.shape

    x = keypoints[..., 0].unsqueeze(1)
    y = keypoints[..., 1].unsqueeze(1)


    gs = generate_gaussian_heatmaps(x, y,num_points=1, heatmap_size=heatmap_size, sigma=1.0)
    # show_heatmap(gs[0],'target')
    all_roi_heatmap = []
    for roi, heatmap in zip(rois, gs):
        # show_heatmap(heatmap, 'target')
        # print(f'heatmap:{heatmap.shape}')
        heatmap = heatmap.unsqueeze(0)
        x1, y1, x2, y2 = map(int, roi)
        roi_heatmap = torch.zeros_like(heatmap)
        roi_heatmap[..., y1:y2 + 1, x1:x2 + 1] = heatmap[..., y1:y2 + 1, x1:x2 + 1]
        # show_heatmap(roi_heatmap[0],'roi_heatmap')
        all_roi_heatmap.append(roi_heatmap)

    all_roi_heatmap = torch.cat(all_roi_heatmap)
    print(f'all_roi_heatmap:{all_roi_heatmap.shape}')

    return all_roi_heatmap

def arc_inference1(arc_equation, x, arc_boxes, th):
    # type: (Tensor, List[Tensor]) -> Tuple[List[Tensor], List[Tensor]]

    points_probs = []
    points_scores = []

    print(f'arc_boxes:{len(arc_boxes)}')

    boxes_per_image = [box.size(0) for box in arc_boxes]
    print(f'arc boxes_per_image:{boxes_per_image}')
    x2 = x.split(boxes_per_image, dim=0)
    arc7 = arc_equation.split(boxes_per_image, dim=0)
    # print(f'arc7:{arc7}')

    for xx, bb in zip(x2, arc_boxes):
        point_prob, point_scores = heatmaps_to_arc(xx, bb)

        points_probs.append(point_prob.unsqueeze(1))
        points_scores.append(point_scores)

    return arc7, points_scores


def points_to_heatmap(keypoints, rois,num_points=2, heatmap_size=(512,512)):
    # type: (Tensor, Tensor, int) -> Tensor
    print(f'rois:{rois.shape}')
    print(f'heatmap_size:{heatmap_size}')


    print(f'keypoints.shape:{keypoints.shape}')
    # batch_size, num_keypoints, _ = keypoints.shape

    x = keypoints[..., 0].unsqueeze(1)
    y = keypoints[..., 1].unsqueeze(1)


    gs = generate_gaussian_heatmaps(x, y,num_points=num_points, heatmap_size=heatmap_size, sigma=2.0)
    # show_heatmap(gs[0],'target')
    all_roi_heatmap = []
    for roi, heatmap in zip(rois, gs):
        # show_heatmap(heatmap, 'target')
        # print(f'heatmap:{heatmap.shape}')
        heatmap = heatmap.unsqueeze(0)
        x1, y1, x2, y2 = map(int, roi)
        roi_heatmap = torch.zeros_like(heatmap)
        roi_heatmap[..., y1:y2 + 1, x1:x2 + 1] = heatmap[..., y1:y2 + 1, x1:x2 + 1]
        # show_heatmap(roi_heatmap[0],'roi_heatmap')
        all_roi_heatmap.append(roi_heatmap)

    all_roi_heatmap = torch.cat(all_roi_heatmap)
    print(f'all_roi_heatmap:{all_roi_heatmap.shape}')

    return all_roi_heatmap

def line_points_to_heatmap(keypoints, rois, heatmap_size):
    # type: (Tensor, Tensor, int) -> Tensor
    print(f'rois:{rois.shape}')
    print(f'heatmap_size:{heatmap_size}')


    print(f'keypoints.shape:{keypoints.shape}')
    # batch_size, num_keypoints, _ = keypoints.shape

    x = keypoints[..., 0]
    y = keypoints[..., 1]

    gs = generate_gaussian_heatmaps(x, y, heatmap_size,num_points=2, sigma=1.0)
    # show_heatmap(gs[0],'target')
    all_roi_heatmap = []
    for roi, heatmap in zip(rois, gs):
        # print(f'heatmap:{heatmap.shape}')
        # show_heatmap(heatmap,'target')
        heatmap = heatmap.unsqueeze(0)

        x1, y1, x2, y2 = map(int, roi)
        roi_heatmap = torch.zeros_like(heatmap)
        roi_heatmap[..., y1:y2 + 1, x1:x2 + 1] = heatmap[..., y1:y2 + 1, x1:x2 + 1]
        # show_heatmap(roi_heatmap[0],'roi_heatmap')
        all_roi_heatmap.append(roi_heatmap)

    if len(all_roi_heatmap) > 0:
        all_roi_heatmap = torch.cat(all_roi_heatmap)
        print(f'all_roi_heatmap:{all_roi_heatmap.shape}')
    else:
        all_roi_heatmap = None

    return all_roi_heatmap


"""
修改适配的原结构的点 转热图，适用于带roi_pool版本的
"""


def line_points_to_heatmap_(keypoints, rois, heatmap_size):
    # type: (Tensor, Tensor, int) -> Tuple[Tensor, Tensor]
    print(f'rois:{rois.shape}')
    print(f'heatmap_size:{heatmap_size}')
    offset_x = rois[:, 0]
    offset_y = rois[:, 1]
    scale_x = heatmap_size / (rois[:, 2] - rois[:, 0])
    scale_y = heatmap_size / (rois[:, 3] - rois[:, 1])

    offset_x = offset_x[:, None]
    offset_y = offset_y[:, None]
    scale_x = scale_x[:, None]
    scale_y = scale_y[:, None]

    print(f'keypoints.shape:{keypoints.shape}')
    # batch_size, num_keypoints, _ = keypoints.shape

    x = keypoints[..., 0]
    y = keypoints[..., 1]

    # gs=generate_gaussian_heatmaps(x,y,512,1.0)

    # print(f'gs_heatmap shape:{gs.shape}')
    #
    # show_heatmap(gs[0],'target')

    x_boundary_inds = x == rois[:, 2][:, None]
    y_boundary_inds = y == rois[:, 3][:, None]

    x = (x - offset_x) * scale_x
    x = x.floor().long()
    y = (y - offset_y) * scale_y
    y = y.floor().long()

    x[x_boundary_inds] = heatmap_size - 1
    y[y_boundary_inds] = heatmap_size - 1
    # print(f'heatmaps x:{x}')
    # print(f'heatmaps y:{y}')

    valid_loc = (x >= 0) & (y >= 0) & (x < heatmap_size) & (y < heatmap_size)
    vis = keypoints[..., 2] > 0
    valid = (valid_loc & vis).long()

    gs_heatmap = generate_gaussian_heatmaps(x, y, heatmap_size, 1.0)

    # show_heatmap(gs_heatmap[0], 'feature')

    # print(f'gs_heatmap:{gs_heatmap.shape}')
    #
    # lin_ind = y * heatmap_size + x
    # print(f'lin_ind:{lin_ind.shape}')
    # heatmaps = lin_ind * valid

    return gs_heatmap


def generate_gaussian_heatmaps(xs, ys, heatmap_size,num_points=2, sigma=2.0, device='cuda'):
    """
    为一组点生成并合并高斯热图。

    Args:
        xs (Tensor): 形状为 (N, 2) 的所有点的 x 坐标
        ys (Tensor): 形状为 (N, 2) 的所有点的 y 坐标
        heatmap_size (int): 热图大小 H=W
        sigma (float): 高斯核标准差
        device (str): 设备类型 ('cpu' or 'cuda')

    Returns:
        Tensor: 形状为 (H, W) 的合并后的热图
    """

    assert xs.shape == ys.shape, "x and y must have the same shape"
    print(f'xs:{xs.shape}')
    xs=xs.squeeze(1)
    ys = ys.squeeze(1)
    print(f'xs1:{xs.shape}')
    N = xs.shape[0]
    print(f'N:{N},num_points:{num_points}')

    # 创建网格
    grid_y, grid_x = torch.meshgrid(
        torch.arange(heatmap_size, device=device),
        torch.arange(heatmap_size, device=device),
        indexing='ij'
    )

    # print(f'heatmap_size:{heatmap_size}')
    # 初始化输出热图
    combined_heatmap = torch.zeros((N, heatmap_size, heatmap_size), device=device)

    for i in range(N):
        heatmap= torch.zeros((heatmap_size, heatmap_size), device=device)
        for j in range(num_points):
            mu_x1 = xs[i, j].clamp(0, heatmap_size - 1).item()
            mu_y1 = ys[i, j].clamp(0, heatmap_size - 1).item()
            # print(f'mu_x1,mu_y1:{mu_x1},{mu_y1}')

            # 计算距离平方
            dist1 = (grid_x - mu_x1) ** 2 + (grid_y - mu_y1) ** 2

            # 计算高斯分布
            heatmap1 = torch.exp(-dist1 / (2 * sigma ** 2))

            heatmap+=heatmap1


        # mu_x2 = xs[i, 1].clamp(0, heatmap_size - 1).item()
        # mu_y2 = ys[i, 1].clamp(0, heatmap_size - 1).item()
        #
        # # 计算距离平方
        # dist2 = (grid_x - mu_x2) ** 2 + (grid_y - mu_y2) ** 2
        #
        # # 计算高斯分布
        # heatmap2 = torch.exp(-dist2 / (2 * sigma ** 2))
        #
        # heatmap = heatmap1 + heatmap2

        # 将当前热图累加到结果中
        combined_heatmap[i] = heatmap

    return combined_heatmap

def generate_mask_gaussian_heatmaps(xs, ys, heatmap_size,num_points=2, sigma=2.0, device='cuda'):
    """
    为一组点生成并合并高斯热图。

    Args:
        xs (Tensor): 形状为 (N, 2) 的所有点的 x 坐标
        ys (Tensor): 形状为 (N, 2) 的所有点的 y 坐标
        heatmap_size (int): 热图大小 H=W
        sigma (float): 高斯核标准差
        device (str): 设备类型 ('cpu' or 'cuda')

    Returns:
        Tensor: 形状为 (H, W) 的合并后的热图
    """

    assert xs.shape == ys.shape, "x and y must have the same shape"
    print(f'xs:{xs.shape}')
    xs=xs.squeeze(1)
    ys = ys.squeeze(1)
    print(f'xs1:{xs.shape}')
    N = xs.shape[0]
    print(f'N:{N},num_points:{num_points}')

    # 创建网格
    grid_y, grid_x = torch.meshgrid(
        torch.arange(heatmap_size, device=device),
        torch.arange(heatmap_size, device=device),
        indexing='ij'
    )

    # print(f'heatmap_size:{heatmap_size}')
    # 初始化输出热图
    combined_heatmap = torch.zeros((N, heatmap_size, heatmap_size), device=device)

    for i in range(N):
        heatmap= torch.zeros((heatmap_size, heatmap_size), device=device)
        for j in range(num_points):
            mu_x1 = xs[i, j].clamp(0, heatmap_size - 1).item()
            mu_y1 = ys[i, j].clamp(0, heatmap_size - 1).item()
            # print(f'mu_x1,mu_y1:{mu_x1},{mu_y1}')

            # 计算距离平方
            dist1 = (grid_x - mu_x1) ** 2 + (grid_y - mu_y1) ** 2

            # 计算高斯分布
            heatmap1 = torch.exp(-dist1 / (2 * sigma ** 2))

            heatmap+=heatmap1


        # mu_x2 = xs[i, 1].clamp(0, heatmap_size - 1).item()
        # mu_y2 = ys[i, 1].clamp(0, heatmap_size - 1).item()
        #
        # # 计算距离平方
        # dist2 = (grid_x - mu_x2) ** 2 + (grid_y - mu_y2) ** 2
        #
        # # 计算高斯分布
        # heatmap2 = torch.exp(-dist2 / (2 * sigma ** 2))
        #
        # heatmap = heatmap1 + heatmap2

        # 将当前热图累加到结果中
        combined_heatmap[i] = heatmap

    return combined_heatmap

def non_maximum_suppression(a):
    ap = F.max_pool2d(a, 3, stride=1, padding=1)
    mask = (a == ap).float().clamp(min=0.0)
    return a * mask


def heatmaps_to_points(maps, rois,num_points=2):


    point_preds = torch.zeros((len(rois),  2), dtype=torch.float32, device=maps.device)
    point_end_scores = torch.zeros((len(rois), 1), dtype=torch.float32, device=maps.device)

    print(f'heatmaps_to_lines:{maps.shape}')
    point_maps=maps[:,0]
    print(f'point_map:{point_maps.shape}')
    for i in range(len(rois)):

        point_roi_map = point_maps[i].unsqueeze(0)
        print(f'point_roi_map:{point_roi_map.shape}')
        # roi_map_probs = scores_to_probs(roi_map.copy())
        w = point_roi_map.shape[2]
        flatten_point_roi_map = non_maximum_suppression(point_roi_map).reshape(1, -1)
        point_score, point_index = torch.topk(flatten_point_roi_map, k=num_points)
        print(f'point index:{point_index}')
        # pos = roi_map.reshape(num_keypoints, -1).argmax(dim=1)

        point_x =point_index % w
        point_y = torch.div(point_index - point_x, w, rounding_mode="floor")
        point_preds[i, 0,] = point_x
        point_preds[i, 1,] = point_y

        point_end_scores[i, :] = point_roi_map[torch.arange(1, device=point_roi_map.device), point_y, point_x]


    return point_preds,point_end_scores

# 分4块
def find_max_heat_point_in_each_part(feature_map, box):
    """
    在给定的特征图上，根据box中心点往上移3，往右移3作为新的中心点，
    并将特征图划分为4个部分，之后在每个部分中找到热度值最大的点。

    Args:
        feature_map (torch.Tensor): 形状为 [C, H, W] 的特征图
        box (torch.Tensor): 形状为 [4] 的边界框 [x_min, y_min, x_max, y_max]

    Returns:
        list: 每个区域中热度最高的点的位置和其对应的热度值 [(y1, x1, heat1), ..., (y4, x4, heat4)]
    """
    device = feature_map.device
    C, H, W = feature_map.shape

    # 计算box的中心点(cx, cy)
    cx = (box[0] + box[2]) // 2
    cy = (box[1] + box[3]) // 2

    # 偏移中心点
    new_cx = min(max(cx + 3, 0), W - 1)  # 向右移3
    new_cy = min(max(cy - 3, 0), H - 1)  # 向上移3

    # 创建坐标网格
    y_coords, x_coords = torch.meshgrid(
        torch.arange(H, device=device), torch.arange(W, device=device), indexing='ij'
    )

    # 划分四个区域
    mask_q1 = (y_coords < new_cy) & (x_coords < new_cx)  # 左上
    mask_q2 = (y_coords < new_cy) & (x_coords >= new_cx)  # 右上
    mask_q3 = (y_coords >= new_cy) & (x_coords < new_cx)  # 左下
    mask_q4 = (y_coords >= new_cy) & (x_coords >= new_cx)  # 右下

    # def process_region(ins):
    #     region = feature_map[:, :, ins].squeeze()
    #     if len(region.shape) == 0:  # 如果区域为空，则跳过
    #         return None, None
    #     # 找到最大热度值的点及其位置
    #     (y, x), heat_val = non_maximum_suppression(region[0])
    #     # 将相对坐标转换回全局坐标
    #     y_global = y + torch.where(ins)[0].min().item()
    #     x_global = x + torch.where(ins)[1].min().item()
    #     return (y_global, x_global), heat_val
    #
    # results = []
    # for ins in [mask_q1, mask_q2, mask_q3, mask_q4]:
    #     point, heat_val = process_region(ins)
    #     if point is not None:
    #         # results.append((point[0], point[1], heat_val))
    #         results.append((point[0], point[1]))
    #     else:
    #         results.append(None)
    masks = [mask_q1, mask_q2, mask_q3, mask_q4]
    results = []

    # 假设使用第一个通道作为热力图
    heatmap = feature_map[0]  # [H, W]

    def process_region(mask):
        # 应用 ins，只保留该区域
        masked_heatmap = heatmap.clone()  # 复制以避免修改原数据
        masked_heatmap[~mask] = 0  # 非区域置0

        def non_maximum_suppression_2d(heatmap, kernel_size=3):
            """
            对 2D 热力图做非极大值抑制，保留局部最大值点。

            Args:
                heatmap (torch.Tensor): [H, W]，输入热力图
                kernel_size (int): 池化窗口大小，用于比较是否为局部最大值

            Returns:
                torch.Tensor: 与 heatmap 同形状的 ins，局部最大值位置为 True
            """
            pad = (kernel_size - 1) // 2
            max_pool = torch.nn.MaxPool2d(kernel_size=kernel_size, stride=1, padding=pad)
            maxima = max_pool(heatmap.unsqueeze(0)).squeeze(0)
            # 局部最大值且值大于0
            peaks = (heatmap == maxima) & (heatmap > 0)
            return peaks

        # 1. 先做 NMS 得到候选局部极大值点
        nms_mask = non_maximum_suppression_2d(masked_heatmap, kernel_size=3)  # [H, W] bool
        candidate_peaks = masked_heatmap * nms_mask.float()  # 只保留 NMS 后的峰值

        # 2. 找出所有候选点中值最大的一个
        if candidate_peaks.max() <= 0:
            return None

        # 找到最大值的位置
        max_val, max_idx = torch.max(candidate_peaks.view(-1), dim=0)
        y, x = divmod(max_idx.item(), W)

        return (x, y)  # 返回 (y, x)

    for mask in masks:
        point = process_region(mask)
        results.append(point)

    return results


def non_maximum_suppression_2d(heatmap, kernel_size=3):
    pad = (kernel_size - 1) // 2
    max_pool = torch.nn.MaxPool2d(kernel_size=kernel_size, stride=1, padding=pad)
    maxima = max_pool(heatmap.unsqueeze(0)).squeeze(0)
    peaks = (heatmap == maxima) & (heatmap > 0)
    return peaks

def find_max_heat_point_in_edge_centers(feature_map, box):

    device = feature_map.device
    C, H, W = feature_map.shape

    # ¼ÆËã box ÖÐÐÄ
    cx = (box[0] + box[2]) / 2
    cy = (box[1] + box[3]) / 2

    # ¸ù¾Ý box ¿í¸ß¼ÆËã¾Å¹¬¸ñ·Ö½çÏß
    box_width = box[2] - box[0]
    box_height = box[3] - box[1]

    x_left = cx - box_width / 6
    x_right = cx + box_width / 6
    y_top = cy - box_height / 6
    y_bottom = cy + box_height / 6

    # ´´½¨Íø¸ñ
    y_coords, x_coords = torch.meshgrid(
        torch.arange(H, device=device),
        torch.arange(W, device=device),
        indexing='ij'
    )

    # ¶¨ÒåËĸö¡°±ßÖС±ÇøÓòµÄ ins
    mask1 = (x_coords < x_left) & (y_coords < y_top)
    mask_top_middle    = (x_coords >= x_left) & (x_coords < x_right) & (y_coords < y_top)
    mask3 = (x_coords >= x_right) & (y_coords < y_top)

    mask_left_middle = (x_coords < x_left) & (y_coords >= y_top) & (y_coords < y_bottom)
    mask_right_middle  = (x_coords >= x_right) & (y_coords >= y_top) & (y_coords < y_bottom)

    mask4 = (x_coords < x_left) & (y_coords >= y_bottom)
    mask_bottom_middle = (x_coords >= x_left) & (x_coords < x_right) & (y_coords >= y_bottom)
    mask_right_bottom = (x_coords >= x_right) & (y_coords >= y_bottom)

    # masks = [
    #     # mask1,
    #     mask_top_middle,
    #     # mask3,
    #     mask_left_middle,
    #     mask_right_middle,
    #     # mask4,
    #     mask_bottom_middle,
    #     mask_right_bottom
    # ]

    masks = [
        mask_top_middle,
        mask_right_middle,
        mask_bottom_middle,
        mask_left_middle
    ]

    # ʹÓõÚÒ»¸öͨµÀ×÷ΪÈÈÁ¦Í¼
    heatmap = feature_map[0]  # [H, W]

    results = []

    for mask in masks:
        masked_heatmap = heatmap.clone()
        masked_heatmap[~mask] = 0  # ·ÇÄ¿±êÇøÓòÖà 0

        # # NMS ÒÖÖÆ
        # nms_mask = non_maximum_suppression_2d(masked_heatmap, kernel_size=3)
        # candidate_peaks = masked_heatmap * nms_mask.float()
        #
        # if candidate_peaks.max() <= 0:
        #     results.append(None)
        #     continue
        #
        # # ÕÒ×î´óֵλÖÃ
        # max_val, max_idx = torch.max(candidate_peaks.view(-1), dim=0)
        # y, x = divmod(max_idx.item(), W)
        flatten_point_roi_map = masked_heatmap.reshape(1, -1)
        point_score, point_index = torch.topk(flatten_point_roi_map, k=1)
        point_x =point_index % W
        point_y = torch.div(point_index - point_x, W, rounding_mode="floor")

        results.append((point_x, point_y))

    return results  # [(y_top, x_top), (y_right, x_right), (y_bottom, x_bottom), (y_left, x_left)]



def heatmaps_to_circle_points(maps, rois,num_points=2):


    point_preds = torch.zeros((len(rois), 4, 2), dtype=torch.float32, device=maps.device)
    point_end_scores = torch.zeros((len(rois),4, 1), dtype=torch.float32, device=maps.device)

    print(f'rois in heatmaps_to_circle_points:{type(rois),  rois.shape}')   # <class 'torch.Tensor'>

    print(f'heatmaps_to_lines:{maps.shape}')
    point_maps=maps[:,0]
    print(f'point_map:{point_maps.shape}')
    for i in range(len(rois)):

        point_roi_map = point_maps[i].unsqueeze(0)
        print(f'point_roi_map:{point_roi_map.shape}')
        # roi_map_probs = scores_to_probs(roi_map.copy())

        # w = point_roi_map.shape[2]
        # flatten_point_roi_map = non_maximum_suppression(point_roi_map).reshape(1, -1)
        # print(f'non_maximum_suppression :{non_maximum_suppression(point_roi_map).shape}')
        # point_score, point_index = torch.topk(flatten_point_roi_map, k=num_points)
        # print(f'point index:{point_index}')
        # point_x =point_index % w
        # point_y = torch.div(point_index - point_x, w, rounding_mode="floor")
        # print(f'point_x:{point_x}, point_y:{point_y}')
        # point_preds[i, :, 0] = point_x
        # point_preds[i, :, 1] = point_y
        roi1=rois[i]
        result_points = find_max_heat_point_in_edge_centers(non_maximum_suppression(point_roi_map), roi1)

        point_preds[i, :]=torch.tensor(result_points)

        point_x = [point[0] for point in result_points]
        point_y = [point[1] for point in result_points]

        point_end_scores[i, :,0] = point_roi_map[torch.arange(1, device=point_roi_map.device), point_y, point_x]


    return point_preds,point_end_scores

def heatmaps_to_lines(maps, rois):
    line_preds = torch.zeros((len(rois), 3, 2), dtype=torch.float32, device=maps.device)
    line_end_scores = torch.zeros((len(rois), 2), dtype=torch.float32, device=maps.device)

    line_maps=maps[:,1]
    # line_maps = maps.squeeze(1)

    for i in range(len(rois)):
        line_roi_map = line_maps[i].unsqueeze(0)

        print(f'line_roi_map:{line_roi_map.shape}')
        # roi_map_probs = scores_to_probs(roi_map.copy())
        w = line_roi_map.shape[1]
        flatten_line_roi_map = non_maximum_suppression(line_roi_map).reshape(1, -1)
        line_score, line_index = torch.topk(flatten_line_roi_map, k=2)
        print(f'line index:{line_index}')
        # pos = roi_map.reshape(num_keypoints, -1).argmax(dim=1)
        pos = line_index
        line_x = pos % w
        line_y = torch.div(pos - line_x, w, rounding_mode="floor")
        line_preds[i, 0, :] = line_x
        line_preds[i, 1, :] = line_y
        line_preds[i, 2, :] = 1
        line_end_scores[i, :] = line_roi_map[torch.arange(1, device=line_roi_map.device), line_y, line_x]




    return line_preds.permute(0, 2, 1), line_end_scores


# 显示热图的函数
def show_heatmap(heatmap, title="Heatmap"):
    """
    使用 matplotlib 显示热图。

    Args:
        heatmap (Tensor): 要显示的热图张量
        title (str): 图表标题
    """
    # 如果在 GPU 上，首先将其移动到 CPU 并转换为 numpy 数组
    if heatmap.is_cuda:
        heatmap = heatmap.cpu().numpy()
    else:
        heatmap = heatmap.numpy()

    plt.imshow(heatmap, cmap='hot', interpolation='nearest')
    plt.colorbar()
    plt.title(title)
    plt.show()

def lines_point_pair_loss(line_logits, proposals, gt_lines, line_matched_idxs):
    # type: (Tensor, List[Tensor], List[Tensor], List[Tensor]) -> Tensor
    N, K, H, W = line_logits.shape
    len_proposals = len(proposals)
    print(f'lines_point_pair_loss line_logits.shape:{line_logits.shape},len_proposals:{len_proposals},line_matched_idxs:{line_matched_idxs}')
    if H != W:
        raise ValueError(
            f"line_logits height and width (last two elements of shape) should be equal. Instead got H = {H} and W = {W}"
        )
    discretization_size = H
    heatmaps = []
    gs_heatmaps = []
    valid = []
    for proposals_per_image, gt_kp_in_image, midx in zip(proposals, gt_lines, line_matched_idxs):
        print(f'line_proposals_per_image:{proposals_per_image.shape}')
        print(f'gt_lines:{gt_lines}')
        if proposals_per_image.shape[0] > 0 and gt_kp_in_image.shape[0] > 0:
            kp = gt_kp_in_image[midx]

            gs_heatmaps_per_img = line_points_to_heatmap(kp, proposals_per_image, discretization_size)
            gs_heatmaps.append(gs_heatmaps_per_img)
        # print(f'heatmaps_per_image:{heatmaps_per_image.shape}')

        # heatmaps.append(heatmaps_per_image.view(-1))

        # valid.append(valid_per_image.view(-1))

    # line_targets = torch.cat(heatmaps, dim=0)
    gs_heatmaps = torch.cat(gs_heatmaps, dim=0)
    print(f'gs_heatmaps:{gs_heatmaps.shape}, line_logits.shape:{line_logits.squeeze(1).shape}')
    # print(f'line_targets:{line_targets.shape},{line_targets}')

    # valid = torch.cat(valid, dim=0).to(dtype=torch.uint8)
    # valid = torch.where(valid)[0]

    # print(f' line_targets[valid]:{line_targets[valid]}')

    # torch.mean (in binary_cross_entropy_with_logits) doesn't
    # accept empty tensors, so handle it sepaartely
    # if line_targets.numel() == 0 or len(valid) == 0:
    #     return line_logits.sum() * 0

    # line_logits = line_logits.view(N * K, H * W)
    # print(f'line_logits[valid]:{line_logits[valid].shape}')
    print(f'loss1 line_logits:{line_logits.shape}')
    line_logits = line_logits[:,1,:,:]
    # line_logits = line_logits.squeeze(1)
    print(f'loss2 line_logits:{line_logits.shape}')

    # line_loss = F.cross_entropy(line_logits[valid], line_targets[valid])
    line_loss = F.cross_entropy(line_logits, gs_heatmaps)

    return line_loss

def compute_ins_loss(feature_logits, proposals, gt_, pos_matched_idxs):
    print(f'compute_arc_loss:{feature_logits.shape}')
    N, K, H, W = feature_logits.shape

    len_proposals = len(proposals)

    empty_count = 0
    non_empty_count = 0

    for prop in proposals:
        if prop.shape[0] == 0:
            empty_count += 1
        else:
            non_empty_count += 1

    print(f"Empty proposals count: {empty_count}")
    print(f"Non-empty proposals count: {non_empty_count}")

    print(f'starte to compute_point_loss')
    print(f'compute_point_loss line_logits.shape:{feature_logits.shape},len_proposals:{len_proposals}')
    if H != W:
        raise ValueError(
            f"line_logits height and width (last two elements of shape) should be equal. Instead got H = {H} and W = {W}"
        )
    discretization_size = H

    gs_heatmaps = []
    # print(f'point_matched_idxs:{point_matched_idxs}')
    print(f'gt_masks:{gt_[0].shape}')
    for proposals_per_image, gt_kp_in_image, midx in zip(proposals, gt_, pos_matched_idxs):
        # [
        #   (Tensor(38, 4), Tensor(1, 57, 2), Tensor(38, 1)),
        #   (Tensor(65, 4), Tensor(1, 74, 2), Tensor(65, 1))
        # ]
        print(f'proposals_per_image:{proposals_per_image.shape}')
        kp = gt_kp_in_image[midx]
        t_h, t_w = kp.shape[-2:]
        print(f't_h:{t_h}, t_w:{t_w}')

        print(f'gt_kp_in_image:{gt_kp_in_image.shape}')
        if proposals_per_image.shape[0] > 0 and gt_kp_in_image.shape[0] > 0:

            gs_heatmaps_per_img = align_masks(kp, proposals_per_image, discretization_size)
            gs_heatmaps.append(gs_heatmaps_per_img)

    if len(gs_heatmaps)>0:
        gs_heatmaps = torch.cat(gs_heatmaps, dim=0)
        print(f'gs_heatmaps:{gs_heatmaps.shape}, line_logits.shape:{feature_logits.shape}')

        line_logits = feature_logits.squeeze(1)
        print(f'ins shape:{line_logits.shape}')

        # line_loss = F.binary_cross_entropy_with_logits(line_logits, gs_heatmaps)

        # line_loss = F.cross_entropy(line_logits, gs_heatmaps)
        line_loss=combined_loss(line_logits, gs_heatmaps)

    else:
        line_loss=100

    print("d")

    return line_loss


def align_masks(keypoints, rois, heatmap_size):
    print(f'rois:{rois.shape}')
    print(f'heatmap_size:{heatmap_size}')

    print(f'keypoints.shape:{keypoints.shape}')
    # batch_size, num_keypoints, _ = keypoints.shape
    t_h, t_w = keypoints.shape[-2:]
    scale=heatmap_size/t_w
    print(f'scale:{scale}')
    x = keypoints[..., 0]*scale
    y = keypoints[..., 1]*scale

    x = x.unsqueeze(1)
    y = y.unsqueeze(1)

    num_points=x.shape[2]
    print(f'num_points:{num_points}')
    mask_4d = keypoints.unsqueeze(1).float()
    resized_mask = F.interpolate(
        mask_4d,
    size = (heatmap_size, heatmap_size),
    mode = 'bilinear',
    align_corners = False
    ).squeeze(1)  # [B,heatmap_size,heatmap_size]
    # plt.imshow(resized_mask[0].cpu())
    # plt.show()
    print(f'resized_mask:{resized_mask.shape}')
    return resized_mask

def compute_point_loss(line_logits, proposals, gt_points, point_matched_idxs):
    # type: (Tensor, List[Tensor], List[Tensor], List[Tensor]) -> Tensor
    N, K, H, W = line_logits.shape
    len_proposals = len(proposals)

    empty_count = 0
    non_empty_count = 0

    for prop in proposals:
        if prop.shape[0] == 0:
            empty_count += 1
        else:
            non_empty_count += 1

    print(f"Empty proposals count: {empty_count}")
    print(f"Non-empty proposals count: {non_empty_count}")

    print(f'starte to compute_point_loss')
    print(f'compute_point_loss line_logits.shape:{line_logits.shape},len_proposals:{len_proposals}')
    if H != W:
        raise ValueError(
            f"line_logits height and width (last two elements of shape) should be equal. Instead got H = {H} and W = {W}"
        )
    discretization_size = H

    gs_heatmaps = []
    # print(f'point_matched_idxs:{point_matched_idxs}')
    for proposals_per_image, gt_kp_in_image, midx in zip(proposals, gt_points, point_matched_idxs):
        print(f'proposals_per_image:{proposals_per_image.shape}')
        kp = gt_kp_in_image[midx]
        # print(f'gt_kp_in_image:{gt_kp_in_image}')
        gs_heatmaps_per_img = single_point_to_heatmap(kp, proposals_per_image, discretization_size)
        gs_heatmaps.append(gs_heatmaps_per_img)

    gs_heatmaps = torch.cat(gs_heatmaps, dim=0)
    print(f'gs_heatmaps:{gs_heatmaps.shape}, line_logits.shape:{line_logits.squeeze(1).shape}')

    line_logits = line_logits[:,0]
    print(f'single_point_logits:{line_logits.shape}')

    line_loss = F.cross_entropy(line_logits, gs_heatmaps)

    return line_loss


def compute_circle_loss(circle_logits, proposals, gt_circles, circle_matched_idxs):
    # type: (Tensor, List[Tensor], List[Tensor], List[Tensor]) -> Tensor
    N, K, H, W = circle_logits.shape
    len_proposals = len(proposals)

    empty_count = 0
    non_empty_count = 0

    for prop in proposals:
        if prop.shape[0] == 0:
            empty_count += 1
        else:
            non_empty_count += 1

    print(f"Empty proposals count: {empty_count}")
    print(f"Non-empty proposals count: {non_empty_count}")

    print(f'starte to compute_circle_loss')
    print(f'compute_circle_loss circle_logits.shape:{circle_logits.shape},len_proposals:{len_proposals}')
    if H != W:
        raise ValueError(
            f"line_logits height and width (last two elements of shape) should be equal. Instead got H = {H} and W = {W}"
        )
    discretization_size = H

    gs_heatmaps = []
    # print(f'point_matched_idxs:{point_matched_idxs}')
    for proposals_per_image, gt_kp_in_image, midx in zip(proposals, gt_circles, circle_matched_idxs):
        print(f'proposals_per_image:{proposals_per_image.shape}')
        kp = gt_kp_in_image[midx]
        # print(f'gt_kp_in_image:{gt_kp_in_image}')
        gs_heatmaps_per_img = points_to_heatmap(kp, proposals_per_image,num_points=4, heatmap_size=discretization_size)
        gs_heatmaps.append(gs_heatmaps_per_img)

    gs_heatmaps = torch.cat(gs_heatmaps, dim=0)
    print(f'gs_heatmaps:{gs_heatmaps.shape}, line_logits.shape:{circle_logits.squeeze(1).shape}')

    circle_logits = circle_logits[:, 0]
    print(f'circle_logits:{circle_logits.shape}')

    circle_loss = F.cross_entropy(circle_logits, gs_heatmaps)

    return circle_loss

def lines_to_boxes(lines, img_size=511):
    """
    输入:
        lines: Tensor of shape (N, 2, 2)，表示 N 条线段，每个线段有两个端点 (x, y)
        img_size: int，图像尺寸，用于 clamp 边界

    输出:
        boxes: Tensor of shape (N, 4)，表示 N 个包围盒 [x_min, y_min, x_max, y_max]
    """
    # 提取所有线段的两个端点
    p1 = lines[:, 0]  # (N, 2)
    p2 = lines[:, 1]  # (N, 2)

    # 每条线段的 x 和 y 坐标
    x_coords = torch.stack([p1[:, 0], p2[:, 0]], dim=1)  # (N, 2)
    y_coords = torch.stack([p1[:, 1], p2[:, 1]], dim=1)  # (N, 2)

    # 计算包围盒边界
    x_min = x_coords.min(dim=1).values
    y_min = y_coords.min(dim=1).values
    x_max = x_coords.max(dim=1).values
    y_max = y_coords.max(dim=1).values

    # 扩展边界并限制在图像范围内
    x_min = (x_min - 1).clamp(min=0, max=img_size)
    y_min = (y_min - 1).clamp(min=0, max=img_size)
    x_max = (x_max + 1).clamp(min=0, max=img_size)
    y_max = (y_max + 1).clamp(min=0, max=img_size)

    # 合成包围盒
    boxes = torch.stack([x_min, y_min, x_max, y_max], dim=1)  # (N, 4)
    return boxes


def box_iou_pairwise(box1, box2):
    """
    输入：
        box1: shape (N, 4)
        box2: shape (M, 4)
    输出：
        ious: shape (min(N, M), ), 只计算 i = j 的配对
    """
    N = min(len(box1), len(box2))
    lt = torch.max(box1[:N, :2], box2[:N, :2])  # 左上角
    rb = torch.min(box1[:N, 2:], box2[:N, 2:])  # 右下角

    wh = (rb - lt).clamp(min=0)  # 宽高
    inter_area = wh[:, 0] * wh[:, 1]  # 交集面积

    area1 = (box1[:N, 2] - box1[:N, 0]) * (box1[:N, 3] - box1[:N, 1])
    area2 = (box2[:N, 2] - box2[:N, 0]) * (box2[:N, 3] - box2[:N, 1])

    union_area = area1 + area2 - inter_area
    ious = inter_area / (union_area + 1e-6)

    return ious


def line_iou_loss(x, boxes, gt_lines, matched_idx, img_size=511, alpha=1.0, beta=1.0, gamma=1.0):
    """
    Args:
        x: [N,1,H,W] 热力图
        boxes: [N,4] 框坐标
        gt_lines: [N,2,3] GT线段（含可见性）
        matched_idx: 匹配 index
        img_size: 图像尺寸
        alpha: IoU 损失权重
        beta: 长度损失权重
        gamma: 方向角度损失权重
    """
    losses = []
    boxes_per_image = [box.size(0) for box in boxes]
    x2 = x.split(boxes_per_image, dim=0)

    for xx, bb, gt_line, mid in zip(x2, boxes, gt_lines, matched_idx):
        p_prob, _ = heatmaps_to_lines(xx, bb)
        pred_lines = p_prob
        gt_line_points = gt_line[mid]

        if len(pred_lines) == 0 or len(gt_line_points) == 0:
            continue

        # IoU 损失
        pred_boxes = lines_to_boxes(pred_lines, img_size)
        gt_boxes = lines_to_boxes(gt_line_points, img_size)
        ious = box_iou_pairwise(pred_boxes, gt_boxes)
        iou_loss = 1.0 - ious  # [N]

        # 长度损失
        pred_len = line_length(pred_lines)
        gt_len = line_length(gt_line_points)
        length_diff = F.l1_loss(pred_len, gt_len, reduction='none')  # [N]

        # 方向角度损失
        pred_dir = line_direction(pred_lines)
        gt_dir = line_direction(gt_line_points)
        ang_loss = angle_loss_cosine(pred_dir, gt_dir)  # [N]

        # 归一化每一项损失
        norm_iou = normalize_tensor(iou_loss)
        norm_len = normalize_tensor(length_diff)
        norm_ang = normalize_tensor(ang_loss)

        total = alpha * norm_iou + beta * norm_len + gamma * norm_ang
        losses.append(total)



    if not losses:
        return None

    return torch.mean(torch.cat(losses))


def point_inference(x, point_boxes):
    # type: (Tensor, List[Tensor]) -> Tuple[List[Tensor], List[Tensor]]

    points_probs = []
    points_scores = []

    boxes_per_image = [box.size(0) for box in point_boxes]
    x2 = x.split(boxes_per_image, dim=0)

    for xx, bb in zip(x2, point_boxes):
        point_prob,point_scores = heatmaps_to_points(xx, bb,num_points=1)

        points_probs.append(point_prob.unsqueeze(1))
        points_scores.append(point_scores)

    return points_probs,points_scores

def circle_inference(x, point_boxes):
    # type: (Tensor, List[Tensor]) -> Tuple[List[Tensor], List[Tensor]]

    points_probs = []
    points_scores = []

    boxes_per_image = [box.size(0) for box in point_boxes]
    x2 = x.split(boxes_per_image, dim=0)

    for xx, bb in zip(x2, point_boxes):
        point_prob,point_scores = heatmaps_to_circle_points(xx, bb,num_points=4)

        points_probs.append(point_prob.unsqueeze(1))
        points_scores.append(point_scores)

    return points_probs,points_scores


def line_inference(x, line_boxes):
    # type: (Tensor, List[Tensor]) -> Tuple[List[Tensor], List[Tensor]]
    lines_probs = []
    lines_scores = []

    boxes_per_image = [box.size(0) for box in line_boxes]
    x2 = x.split(boxes_per_image, dim=0)
    # x2:tuple 2 x2[0]:[1,3,1024,1024]
    # line_box: list:2 [1,4] [1.4] fasterrcnn kuang
    for xx, bb in zip(x2, line_boxes):
        line_prob, line_scores, = heatmaps_to_lines(xx, bb)
        lines_probs.append(line_prob)
        lines_scores.append(line_scores)

    return lines_probs, lines_scores



def ins_inference(x, arc_boxes, th):
    # type: (Tensor, List[Tensor]) -> Tuple[List[Tensor], List[Tensor]]

    points_probs = []
    points_scores = []

    print(f'arc_boxes:{len(arc_boxes)}')

    boxes_per_image = [box.size(0) for box in arc_boxes]
    print(f'arc boxes_per_image:{boxes_per_image}')
    x2 = x.split(boxes_per_image, dim=0)

    for xx, bb in zip(x2, arc_boxes):
        point_prob,point_scores = heatmaps_to_arc(xx, bb)

        points_probs.append(point_prob.unsqueeze(1))
        points_scores.append(point_scores)


    points_probs_tensor=torch.cat(points_probs)
    print(f'points_probs shape:{points_probs_tensor.shape}')

    feature_logits = x
    batch_size = feature_logits.shape[0]
    num_proposals = len(arc_boxes[0])

    results = [[torch.empty(0, 2) for _ in range(num_proposals)] for _ in range(batch_size)]
    proposals_list = arc_boxes[0]  # [[tensor(...)]]

    for proposal_idx, proposal in enumerate(proposals_list):
        coords = proposal.tolist()
        x1, y1, x2, y2 = map(int, coords)

        x1 = max(0, x1)
        y1 = max(0, y1)
        x2 = min(feature_logits.shape[3], x2)
        y2 = min(feature_logits.shape[2], y2)

        for batch_idx in range(batch_size):
            region = feature_logits[batch_idx, :, y1:y2, x1:x2]
            mask = region > th
            coords = torch.nonzero(mask)

            if coords.numel() > 0:
                # 取 (y, x)，然后转换为全局坐标 (x, y)
                local_coords = coords[:, [2, 1]]  # (x, y)
                local_coords[:, 0] += x1
                local_coords[:, 1] += y1

                results[batch_idx][proposal_idx] = local_coords

    print(f're:{results}')

    return points_probs,points_scores,results


import torch.nn.functional as F




def heatmaps_to_arc(maps, rois, threshold=0.5, output_size=(128, 128)):
    """
    Args:
        maps: [N, 3, H, W] - full heatmaps
        rois: [N, 4] - bounding boxes
        threshold: float - binarization threshold
        output_size: resized size for uniform NMS

    Returns:
        masks: [N, 1, H, W] - binary ins aligned with input map
        scores: [N, 1] - count of non-zero pixels in each ins
    """
    N, _, H, W = maps.shape
    masks = torch.zeros((N, 1, H, W), dtype=torch.float32, device=maps.device)
    scores = torch.zeros((N, 1), dtype=torch.float32, device=maps.device)

    point_maps = maps[:, 0]  # È¡µÚÒ»¸öͨµÀ [N, H, W]

    print(f"==> heatmaps_to_arc: maps.shape = {maps.shape}, rois.shape = {rois.shape}")

    for i in range(N):
        x1, y1, x2, y2 = rois[i].long()
        x1 = x1.clamp(0, W - 1)
        x2 = x2.clamp(0, W - 1)
        y1 = y1.clamp(0, H - 1)
        y2 = y2.clamp(0, H - 1)

        print(f"[{i}] roi: ({x1.item()}, {y1.item()}, {x2.item()}, {y2.item()})")

        if x2 <= x1 or y2 <= y1:
            print(f"    Skipped invalid ROI at index {i}")
            continue

        roi_map = point_maps[i, y1:y2, x1:x2]  # [h, w]
        print(f"    roi_map.shape: {roi_map.shape}")

        if roi_map.numel() == 0:
            print(f"    Skipped empty ROI at index {i}")
            continue

        # resize to uniform size
        roi_map_resized = F.interpolate(
            roi_map.unsqueeze(0).unsqueeze(0),
            size=output_size,
            mode='bilinear',
            align_corners=False
        )  # [1, 1, H, W]
        print(f"    roi_map_resized.shape: {roi_map_resized.shape}")

        # NMS + threshold
        # nms_roi = non_maximum_suppression(roi_map_resized)  # shape: [1, H, W]
        nms_roi = torch.sigmoid(roi_map_resized)

        bin_mask = (nms_roi >= threshold).float()  # shape: [1, H, W]
        print(f"    bin_mask.sum(): {bin_mask.sum().item()}")

        # resize back to original roi size
        h = int((y2 - y1).item())
        w = int((x2 - x1).item())
        # È·±£ bin_mask ÊÇ [1, 128, 128]
        assert bin_mask.dim() == 4, f"Expected 3D tensor [1, H, W], got {bin_mask.shape}"
        bin_mask_original_size = F.interpolate(
            bin_mask,  # [1, 1, 128, 128]
            size=(h, w),
            mode='bilinear',
            align_corners=False
        )[0]  # [1, h, w]

        bin_mask_original_size = bin_mask_original_size[0]  # [h, w]

        masks[i, 0, y1:y2, x1:x2] = bin_mask_original_size
        scores[i] = bin_mask_original_size.sum()
        #
        # bin_mask_original_size = F.interpolate(
        #     # bin_mask.unsqueeze(0),  # ? [1, 1, 128, 128]
        #     bin_mask,  # ? [1, 1, 128, 128]
        #     size=(h, w),
        #     mode='bilinear',
        #     align_corners=False
        # )[0]  # ? [1, h, w]
        #
        # masks[i, 0, y1:y2, x1:x2] = bin_mask_original_size.squeeze()
        # scores[i] = bin_mask_original_size.sum()


        # plt.figure(figsize=(6, 6))
        # plt.imshow(masks[i, 0].cpu().numpy(), cmap='gray')
        # plt.title(f"Mask {i}, score={scores[i].item():.1f}")
        # plt.axis('off')
        # plt.show()

        print(f"    bin_mask_original_size.shape: {bin_mask_original_size.shape}, sum: {scores[i].item()}")

    print(f"==> Done. Total valid masks: {(scores > 0).sum().item()} / {N}")

    return masks, scores