Skip to content

Latest commit

 

History

History

scripts

README.md

Scripts

run_pipeline.py

This script creates and trains a pipeline (SemanticSegmentation or ObjectDetection). To define the dataset you can pass the path to the dataset or the path to a config file as shown below.

# Initialize a dataset using its path
python scripts/run_pipeline.py {tf|torch} -p PIPELINE_NAME -m MODEL_NAME \
-d DATASET_NAME --dataset_path DATASET_PATH [optional arguments]

# Initialize a dataset using its config file
python scripts/run_pipeline.py {tf|torch} -p PIPELINE_NAME -m MODEL_NAME \
-d DATASET_NAME --cfg_dataset DATASET_CONFIG_FILE  [optional arguments]

Alternatively, you can run the script using one single config file, which contains configs for dataset, model, and pipeline.

python scripts/run_pipeline.py {tf|torch} -c CONFIG_FILE [optional arguments]

For further help, run python scripts/run_pipeline.py --help.

Examples

# Training on RandLANet and SemanticKITTI with torch.
python scripts/run_pipeline.py torch -c ml3d/configs/randlanet_semantickitti.yml --dataset.dataset_path <path-to-dataset> --pipeline SemanticSegmentation --dataset.use_cache True

# Training on PointPillars and KITTI with torch.
python scripts/run_pipeline.py torch -c ml3d/configs/pointpillars_kitti.yml --split test --dataset.dataset_path <path-to-dataset> --pipeline ObjectDetection --dataset.use_cache True

# Use a config file to train this model with tensorflow
python scripts/run_pipeline.py tf -c ml3d/configs/kpconv_semantickitti.yml \
--dataset_path ../--pipeline.batch_size 2

Arguments can be

  • -p, --pipeline: pipeline name, SemanticSegmentation or ObjectDetection
  • -m, --model: model name (RnadLANet, KPConv)
  • -d, --dataset: dataset name (SemanticKITTI, Toronto3D, S3DIS, ParisLille3D, Semantic3D)
  • -c, --c: config file path (example config files are in in ml3d/configs/)
  • --cfg_model: path to the model's config file
  • --cfg_pipeline: path to the pipeline's config file
  • --cfg_dataset: path to the dataset's config file
  • --cfg_model: path to the model's config file
  • --dataset_path: path to the dataset
  • --device: cpu or gpu

You can also add arbitrary arguments in the command line and the arguments will save in a dictionary and merge with dataset/model/pipeline's existing cfg. For example, --foo abc will add {"foo": "abc"}to the cfg dict.