Make data set from the ground truth.
- Decode the KITTI ground truth result.
- Cut the segment of cloud that contains each object -> to make an object classification dataset.
- Pretreatment of the raw data, and build a PointNet++ to classify them. (Classfication precision reaches 90%) Mine model could be found here in Baidu Yun with extraction code '5u6q' .
- Implement the model to the whole KITTI set for detection task. (Traditional segmentation + Pointnet++ classification)
- Analysis the results, show the recall and precision.
- What todo in the future.