Track 4 is the 3D Point Cloud Classification track. The goal is to classify (semantically segment) point clouds on a per point basis. The classes are:
| Class Index | Class Description |
|---|---|
| 2 | Ground |
| 5 | High Vegetation |
| 6 | Building |
| 9 | Water |
| 17 | Bridge Deck |
For the baseline algorithm, a PointNet++ model was updated with modifications to support splitting/recombining large scenes.
PointSIFT is a semantic segmentation framework for 3D point clouds. It is based on a simple module which extract featrues from neighbor points in eight directions.
Confusion matrix with overall accuracy: 98.33%
T P(2) P(5) P(6) P(9) P(17)
---------------------------------------------------
2 5103290| 65| 20517| 4408| 3066|
5 1810| 1343702| 28977| 2| 4|
6 47554| 16342| 1322237| 0| 1036|
9 3379| 0| 0| 191191| 20|
17 5130| 0| 692| 0| 83878|
mIoU: 0.944256680661
IoU:
Class 2 ( Ground ): 0.9833
Class 5 ( High Vegetation ): 0.9655
Class 6 ( Building ): 0.9184
Class 9 ( Water ): 0.9605
Class 17 ( Elevated Road ): 0.8936
Original repository
Data Augumentation
$ cd utils
$ please refer to utils/README.md
Date Preparation
$ cd dfc
$ python create_train_dataset.py --help
Train & Eval
$ chmod 777 run_5_fold.sh
$ ./run_5_fold.sh
Visualization
We have provided a handy point cloud visualization tool under utils. Run sh compile_render_balls_so.sh to compile it and then you can try the demo with python show3d_balls.py.