JeffWang987 / OpenOccupancy

[ICCV 2023] OpenOccupancy: A Large Scale Benchmark for Surrounding Semantic Occupancy Perception
Apache License 2.0
592 stars 50 forks source link

请问下nuscenes-occupancy数据集的坐标参考系 #50

Open freezing-index opened 6 months ago

freezing-index commented 6 months ago

您好,

请问下 nuscenes-occupancy 数据集中的点云坐标是经过体素化处理后的吗?我直接使用里面的坐标转换到全局坐标时,发现与实际的 nuscenes 数据集相差很大。在应用了下述公式转换后,再转换到全局坐标的效果才差不多: `lidar_rec = nusc.get('sample_data', base_filename) data = data1[:, [2, 1, 0, 3]] tmp = data[:, 0:3].copy() tmp = tmp.T

voxel_size = np.array([0.2, 0.2, 0.2]) # voxel_size 的维度是 (3,) range_min = np.array([-51.2, -51.2, -5]) # range_min 的维度是 (3,)

lidar_coords = tmp * voxel_size[:, np.newaxis] + range_min[:, np.newaxis] + voxel_size[:, np.newaxis] / 2 `

freezing-index commented 6 months ago

2df0f8f0f3868cb3e6a0455bb22a136

freezing-index commented 6 months ago

右上那块大的是没有经过lidar_coords = tmp * voxel_size[:, np.newaxis] + range_min[:, np.newaxis] + voxel_size[:, np.newaxis] / 2转换的,转换后的与nuscenes的在左下角,我不知道我这样转换的是否正确

Agito555 commented 5 months ago

You are right. You can check that the type of gt you load is "int"......