Open zhanghui75 opened 1 week ago
Hi, here are several key points:
vehicles.static is not a vehicle actor (line 1022). Instead, we retrieve it using the API CarlaDataProvider.get_world().get_actors('static.prop.mesh') (line 1031).
The bounding boxes for vehicles.static obtained with get_actors('static.prop.mesh') were found to be inaccurate after visualization.
We used the API get_level_bbs to get bounding boxes. However, this API only returns bounding boxes and cannot provide other information about the vehicles, nor can it achieve one-to-one correspondence.
Therefore, we implemented simple distance matching to establish a one-to-one correspondence.
In you data_collect.py, I found that the bbox of static vehicles are determined by two API calls, and the match between them are set up, like the following (line 1254, data_collect.py).
I found that a quite loose threshold 20 is used (i.e. 'if min_dis > 20:' as below), implying that the two bbox results (i.e. vehicles.static, and vehicles_static_bbox) are quite different. Do you know why the bbox locations between these two results are quite different? Should some transformation be applied to align them?
Furthermore, it seems that get_level_bbs API can't obtain bounding boxes for all vehicles.statc, in my experiments (i.e. the number of returned bboxes from get_level_bbs is much less than the number of vehicles.static). Do you also have such observations? Is there some solution to this problem?
here is your code in data_collect.py
vehicles.static