Semantic segmentation

Here we define the LiDAR semantic segmentation task on S.MID. The goal of this task is to predict the category of every point in each single scan point clouds. There are 14 categories and learning map can be found in our project page . We take mIoU as the evaluation metric.

The baseline provided in the paper are trained and tested based on SMID_beta_v1_2. More details are provided in issue 1.

We are considering raise a competition on test set (S.MID 1.3) in the future.