BEVBox import error

[MariaLyssenko]

Hi,

there is a problem with the BEVBox import.

By calling from zod.visualization.lidar_bev import BEVBox the following error occurs:

mutable default <class 'numpy.ndarray'> for field grid_min is not allowed: use default_factory

Are you able to import the class successfully?

Thanks!

[Petros626]

@MariaLyssenko hey can you tell me how long it normally took to get access to the dataset. Maybe then I could help you with your problem.

[MariaLyssenko]

For me the access request was processed quite fast @Petros626

[Petros626]

@MariaLyssenko days/weeks? Did you used an university email or your private one?

[MariaLyssenko]

days + I used the company email

[Petros626]

Okay then I should use this too. Nobody replied to my request.

Another question would be, if you know the sensor setup heights. For the lidar a sensor height would be beneficial, bc otherwise we don't know the z-axis area.

[MariaLyssenko]

I haven't used those details yet. It might be mentioned in the paper however

[Petros626]

No it's not provided in the paper.

[MariaLyssenko]

the you need to approximate it using the vehicles parameters + the extrinsics

[Petros626]

You can check here: #64 (comment), but Idk why extrinsics should be necessary. I'm only interested in LiDAR height. The problem is the unknown white platform.

[MariaLyssenko]

base coordinate system is the rear axle of the car. with the car geometry that can be derived from the car's model + extrinsic of the lidar you can get a rough estimate of the z-position. However, you might ask the owners of that repo for more insight

[Petros626]

Okay so you mean I have to estimate it for every sample? How would you do that technically?

Btw am I right with my assumption, that they used a right handed coordinate system for the LiDAR just with different orientation of the axis? Normally I know the LiDAR coordinate system to be in same direction like OxTS on their image.

[Petros626]

@MariaLyssenko can you guide me here?

[Petros626]

@MariaLyssenko to come back to your question, for me it was running without problems. But with Python 3.11 and Numpy 2.0 I experienced the same error like you.

Update Solution:
You need to change /home/rlab10/anaconda3/envs/pcdet/lib/python3.11/site-packages/zod/visualization/bev_utils.py:

# General settings
#grid_min: np.ndarray = np.array([-50.0, 0.0])
grid_min: np.ndarray = field(default_factory=lambda: np.array([-50.0, 0.0]))
#grid_max: np.ndarray = np.array([50.0, 100.0])
grid_max: np.ndarray = field(default_factory=lambda: np.array([50.0, 100.0]))

If you got Numpy Version 2.x, replace every np.cast with np.asarray:

point_indices_c = np.asarray((points[:, 2] - settings.pixor_z_min) / settings.grid_cell_size, dtype="int32")
indices_xy = np.asarray((cloud[:, :2] - settings.grid_min) / settings.grid_cell_size, dtype="int32")

In lidar_bev.py replace also np.cast with np.asarray

arrow_length = int(np.asarray((dimension[0]) * 0.8, dtype="float32"))
occupancy = np.maximum.reduce(np.asarray(np.abs(input_array) > 0.0, dtype="float32"), axis=0, keepdims=True)

Which Numpy version gets installed with zod default? How about an PR for this error? @wljungbergh