track4

Track 4: 3D Point Cloud Classification

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

Additionally, some of the ground truth points are marked with a 0 class. This represents unlabeled data, and points with this label will be ignored for metrics purposes.

Baseline Algorithm

For the baseline algorithm, a PointNet++ (aka PointNet2) model was updated with modifications to support splitting/recombining large scenes. For details on setting up/running the model, see pointnet2/dfc/README.md

Model weights

The model weights are now in a GitHub release zip file for download to avoid having large files in the code repo.

Metrics

To run the metrics code, it is easiest to use the same docker container that is used for the model, though it is not necessary. Example command:

docker run -it --rm \
    -v /path/to/data:/data \
    -v /path/to/metrics_code_folder:/metrics \
    dfc_pointcloud bash -c \
    "python /metrics/track4-metrics.py -g /data/ground_truth -d /data/output_data | tee /data/output_data/metrics.txt"