Fast kitti object detection eval in python(finish eval in less than 10 second), only support 2d/bev/3d eval for now.
Only support Anaconda with python 3.6+, need numpy, skimage, numba, fire and cudatoolkit
conda install -c numba cudatoolkit=x.x (depend on your cuda version)
- command interface:
python evaluate.py evaluate --label_path=/path/to/your_gt_label_folder --result_path=/path/to/your_result_folder --label_split_file=/path/to/val.txt --current_class=0
- python interface:
import kitti_common as kitti
from eval import get_official_eval_result
def _read_imageset_file(path):
with open(path, 'r') as f:
lines = f.readlines()
return [int(line) for line in lines]
det_path = "/path/to/your_result_folder"
dt_annos = kitti.get_label_annos(det_path)
gt_path = "/path/to/your_gt_label_folder"
gt_split_file = "/path/to/val.txt" # from https://xiaozhichen.github.io/files/mv3d/imagesets.tar.gz
val_image_ids = _read_imageset_file(gt_split_file)
gt_annos = kitti.get_label_annos(gt_path, val_image_ids)
print(get_official_eval_result(gt_annos, dt_annos, 0))