Questions about the height (z axis) of 3D lane

[qiaozhijian]

I found that the height of 3d lanes might be a little buggy. It is somewhat discontinuous and appears inaccurate. You can see the example as follows. Different views are presented.

However, if I project them to 2D space by removing z axis, they will look good.

Could you please tell me how the height is obtained? By lidar?

How can I process the lanes? Only use XY?

[James-Hays]

Hi @qiaozhijian

It's definitely possible that there are issues with the data, but it looks like the axes you are using to visualize the data are vastly different scales. The x and y dimensions span 60 or 70 meters but the height dimension spans only 60 centimeters. So maybe that is correct and this scene is fairly flat?

[qiaozhijian]

Hi @qiaozhijian

It's definitely possible that there are issues with the data, but it looks like the axes you are using to visualize the data are vastly different scales. The x and y dimensions span 60 or 70 meters but the height dimension spans only 60 centimeters. So maybe that is correct and this scene is fairly flat?

Thank you for your reply. Indeed, each axis has a different scale due to the default settings of matplotlib. But let's take a closer look at the absolute scale, the lanes have a height change of nearly 1m in a local area, which is obviously unreasonable. Of course, if the scale is uniform, then the scene will look flat.

[qiaozhijian]

I noticed that av2 also provided rasterized map of ground height, I thought it would be better to use that to correct the lanes.

[benjaminrwilson]

Hi @qiaozhijian,

Would you mind listing any problematic log ids and we can take a look?

Thanks!

Ben

[qiaozhijian]

Hi @benjaminrwilson, I'm glad to provide the whole visualization code.

from pathlib import Path
import matplotlib.pyplot as plt
import numpy as np
from av2.datasets.sensor.av2_sensor_dataloader import AV2SensorDataLoader
from av2.map.map_api import ArgoverseStaticMap, LaneSegment

if __name__ == '__main__':
    # path to where the logs live
    dataroot = Path("/media/qzj/Extreme SSD/datasets/argoverse/tbv/logs/")
    # unique log identifier
    log_id = "2S1xkgWyVhoa5zwPLz9tl2MAG0lMxj6i__Autumn_2020"

    loader = AV2SensorDataLoader(data_dir=dataroot, labels_dir=dataroot)
    log_map_dirpath = Path(dataroot) / log_id / "map"
    avm = ArgoverseStaticMap.from_map_dir(log_map_dirpath, build_raster=False)

    # retain every pose first.
    traj_ns = loader.get_subsampled_ego_trajectory(log_id, sample_rate_hz=1e9)
    med_x, med_y = np.median(traj_ns, axis=0) # example 1
    # med_x, med_y = traj_ns[10]    # example 2

    # Derive plot area from trajectory (with radius defined in infinity norm).
    view_radius_m = 20
    xlims = [med_x - view_radius_m, med_x + view_radius_m]
    ylims = [med_y - view_radius_m, med_y + view_radius_m]

    fig = plt.figure()
    ax = fig.add_subplot(111, projection='3d')
    vector_lane_segments = avm.get_nearby_lane_segments(query_center=np.asarray([med_x, med_y]), search_radius_m=view_radius_m)
    for vls in vector_lane_segments:
        left_lane_markings = vls.left_lane_marking.polyline
        right_lane_markings = vls.right_lane_marking.polyline
        ax.plot(left_lane_markings[:, 0], left_lane_markings[:, 1], left_lane_markings[:, 2], linewidth=1)
        ax.plot(right_lane_markings[:, 0], right_lane_markings[:, 1], right_lane_markings[:, 2], linewidth=1)

    plt.xlim(*xlims)
    plt.ylim(*ylims)
    plt.legend()
    plt.tight_layout()
    plt.show()