Lane segment data for subset B

[anhtr-nguyn]

Dear Authors,

As in LaneSegNet openreview, I have noticed that you have the pre-processed version for lane segment of subset B. Could you release it ?

[sephyli]

We mentioned that "This benchmark includes only centerline and lane graph annotations. To address this, we produced the pseudo-label of lane segments by assigning a standardized lane width to the lane centerlines".
You can easily obtain the label by transform the points in Frenet coordinates to Cartesian coordinates, with the centerline label in subset B as the reference path.

[anhtr-nguyn]

Hi @sephyli,
According to your openreview reply, in case you keep the script to generate the pseudo-label, could you upload it ? Thank you very much

If not, could you please explain in detail about how to obtain the area label (pedestrian), the class of right line and left line ?

[sephyli]

For subset-B, we only generate pseudo-label of lane segment and evaluate the centerline performance. I found the script, but it can not run directly. You may put it in the dataset, and you may can run it smoothly. I do not guarantee anything for the script.

def gen_pseudo_laneseg(self, index):
    olv2_ann_info = self.data_infos[index]['annotation']
    lane_segments = []
    for lane in olv2_ann_info['lane_centerline']:
        centerline = fix_pts_interpolate(lane['points'], 50)
        whole_direction = centerline[-1] - centerline[0]
        whole_direction = (whole_direction / np.linalg.norm(whole_direction))
        whole_orthogonal_direction = np.cross(whole_direction, np.array([0, 0, 1]))
        if np.dot(whole_orthogonal_direction, np.array([0, 1, 0])) < 0:
            whole_orthogonal_direction = -whole_orthogonal_direction
        whole_orthogonal_direction = whole_orthogonal_direction / np.linalg.norm(whole_orthogonal_direction)
        left_boundary = []
        right_boundary = []
        for i in range(len(centerline)-1):
            direction = centerline[i+1] - centerline[i]
            direction = (direction / np.linalg.norm(direction))
            orthogonal_direction = np.cross(direction, np.array([0, 0, 1]))
            # if np.dot(orthogonal_direction, np.array([0, 1, 0])) < 0:
            if np.dot(orthogonal_direction, whole_orthogonal_direction) < 0:
                orthogonal_direction = -orthogonal_direction
            orthogonal_direction = orthogonal_direction / np.linalg.norm(orthogonal_direction)
            left_boundary.append(centerline[i] + orthogonal_direction * 2)
            right_boundary.append(centerline[i] - orthogonal_direction * 2)
            if i == len(centerline)-1:
                left_boundary.append(centerline[i+1] + orthogonal_direction * 2)
                right_boundary.append(centerline[i+1] - orthogonal_direction * 2)
        left_boundary = np.array(left_boundary)
        right_boundary = np.array(right_boundary)

        lane_segments.append(dict(
            centerline=centerline,
            left_laneline=left_boundary,
            right_laneline=right_boundary,
            left_laneline_type=0,
            right_laneline_type=0,))

    pseudo_laneseg_info = dict(
        lane_segment=lane_segments,
        area=[],
        topology_lsls=olv2_ann_info['topology_lclc'],
    )
    return pseudo_laneseg_info