PolyLaneNet

SCNN、Line-CNN、FastDraw

https://github.com/lucastabelini/PolyLaneNet

https://arxiv.org/pdf/2004.10924.pdf

模型预测输出

前面4个a表示多项式系数

the vertical position h of the horizon line,
which helps to define the upper limit of the lane markings

c车道线得分，每条车道线都有一个。

论文核心

 

target-label

 类别、s、h、车道线上点坐标(xs,ys)

损失代码

        # classification loss
        if self.pred_category and self.extra_outputs > 0:
            ce = nn.CrossEntropyLoss()
            pred_categories = extra_outputs.reshape(target.shape[0] * target.shape[1], -1)
            target_categories = target_categories.reshape(pred_categories.shape[:-1]).long()
            # 只会计算target_categories>0的线
            pred_categories = pred_categories[target_categories > 0]
            target_categories = target_categories[target_categories > 0]
            cls_loss = ce(pred_categories, target_categories - 1)
        else:
            cls_loss = 0

# poly loss calc
        target_xs = target_points[valid_lanes_idx_flat, :target_points.shape[1] // 2]
        ys = target_points[valid_lanes_idx_flat, target_points.shape[1] // 2:].t()
        valid_xs = target_xs >= 0
        pred_polys = pred_polys[valid_lanes_idx_flat]
        
        #利用pre多项式系数，计算坐标
        pred_xs = pred_polys[:, 0] * ys**3 + pred_polys[:, 1] * ys**2 + pred_polys[:, 2] * ys + pred_polys[:, 3]
        pred_xs.t_()
        weights = (torch.sum(valid_xs, dtype=torch.float32) / torch.sum(valid_xs, dim=1, dtype=torch.float32))**0.5
        pred_xs = (pred_xs.t_() *
                   weights).t()  # without this, lanes with more points would have more weight on the cost function
        target_xs = (target_xs.t_() * weights).t()
        poly_loss = mse(pred_xs[valid_xs], target_xs[valid_xs]) / valid_lanes_idx.sum()
        poly_loss = threshold(
            (pred_xs[valid_xs] - target_xs[valid_xs])**2).sum() / (valid_lanes_idx.sum() * valid_xs.sum())

 将预测的多项式系数和target的一个坐标ys相乘，得到预测坐标pred_xs

计算pred_xs和target_xs的mse损失