search

thilius / 3D_BBOX_from_2D

Estimate the 3D bounding box from the detected 2D bounding box
8 stars 5 forks source link

In terms of algorithm implementation, what are the similarities and differences between the Boxy dataset and the Kitti dataset #1

Closed WSTao closed 3 years ago

WSTao commented 3 years ago

The Boxy dataset is a pseudo 3D annotation. The 3D box of the kitti data set is obtained through lidar labeling box mapping. The 3D boxes obtained by these two Datasets are different, so how to deal with the algorithm

thilius commented 3 years ago

@WSTao Both the algorithms have their own train.py. The BOXY dataset should firstly be transfromed with the boxy2dataset.py. Then train with the train.py.

Regarding the different annotation, beside the pseudo annotation, the boxy dataset has only the annotation of the vehicles in the same direction.

The 3D bbox prediction is based on the 2D object detection algorithm, in bbox_3d_prediction.py the YOLOv4 darknet is used.

WSTao commented 3 years ago

OK, Thanks