SSD 复现过程报错

错误：

ImportError: cannot import name 'VoxelGenerator' from 'spconv.utils

官方给的解释

• VoxelGenerator has been replaced by spconv.pytorch.utils.PointToVoxel (torch API) or Point2VoxelGPU[1-4]d/Point2VoxelCPU[1-4]d (tv.Tensor API).

  网络上有很多的提到该错误但并没有彻底去解决，都是copy党

• 笔者给出两种解决办法

• 办法1：

• 将from spconv.utils import VoxelGeneratorV2 as VoxelGenerator

• 改为

•  import spconv.pytorch as spconv
   from spconv.utils import Point2VoxelCPU3d as VoxelGenerator

•    def transform_points_to_voxels(self, data_dict=None, config=None, voxel_generator=None):
          if data_dict is None:
              try:
                  #from spconv.utils import VoxelGeneratorV2 as VoxelGenerator
                  import spconv.pytorch as spconv
                  from spconv.utils import Point2VoxelCPU3d as VoxelGenerator
                  #from spconv.pytorch.utils import PointToVoxel
              except:
                  from spconv.utils import Point2VoxelCPU3d as VoxelGenerator
                  #from spconv.utils import VoxelGenerator
                  #vsize_xyz: List[float[3]], coors_range_xyz: List[
                      #float[6]], num_point_features: int, max_num_voxels: int, max_num_points_per_voxel: int
              voxel_generator = VoxelGenerator(
                  vsize_xyz=config.VOXEL_SIZE,
                  coors_range_xyz=self.point_cloud_range,
                  max_num_points_per_voxel=config.MAX_POINTS_PER_VOXEL,
                  num_point_features=config.NUM_POINT_FEATURES,
                  max_num_voxels=config.MAX_NUMBER_OF_VOXELS[self.mode]
              )
              grid_size = (self.point_cloud_range[3:6] - self.point_cloud_range[0:3]) / np.array(config.VOXEL_SIZE)
              self.grid_size = np.round(grid_size).astype(np.int64)
              self.voxel_size = config.VOXEL_SIZE
              return partial(self.transform_points_to_voxels, voxel_generator=voxel_generator)

  提醒一句，新手的请注意

• num_point_features=config.NUM_POINT_FEATURES, 这一行代码再原yaml文件里，是有这个数据的，把它复制一下在yaml里面就可以了

• 

• 办法2：

• 将from spconv.utils import VoxelGeneratorV2 as VoxelGeneratorfrom

  改为

  spconv.pytorch.utils import PointToVoxel

  二者效果一致，都可以使用

• 另外，笔者配置SSD环境sponcv为2.3.6；python 为3.7；