GETTING_STARTED.md

Getting Started

Please see Getting Started with OpenPCDet for full usage.

Preparing Datasets for M3DETR

Expected dataset structure for Waymo Open Dataset

M3DETR
├── data
│   ├── waymo
│   │   │── ImageSets
│   │   │── raw_data
│   │   │   │── segment-xxxxxxxx.tfrecord
|   |   |   |── ...
|   |   |── waymo_processed_data_v0_5_0
│   │   │   │── segment-xxxxxxxx/
|   |   |   |── ...
│   │   │── waymo_processed_data_v0_5_0_gt_database_train_sampled_1/
│   │   │── waymo_processed_data_v0_5_0_waymo_dbinfos_train_sampled_1.pkl
│   │   │── waymo_processed_data_v0_5_0_gt_database_train_sampled_1_global.npy (optional)
│   │   │── waymo_processed_data_v0_5_0_infos_train.pkl (optional)
│   │   │── waymo_processed_data_v0_5_0_infos_val.pkl (optional)
├── pcdet
├── tools

The dataset configs are located within tools/cfgs/dataset_configs, and the model configs are located within tools/cfgs/m3detr_models for different datasets.

Inference with Pre-trained Models

1. Pick a configuration file and its corresponding model from model zoo, for example, the config file of ./tools/cfgs/m3detr_models/m3detr_waymo_1500.yaml.

2. We provide ./scripts/dist_test.sh that is able to test with a pretrained model. Run it with:

cd tools/
sh ./scripts/dist_test.sh 1 --cfg_file ./cfgs/m3detr_models/m3detr_waymo_1500.yaml --workers 4 --ckpt /path/to/checkpoint_file --eval_tag test_out --batch_size 8 --save_to_file [--other-options]

Train a Model

We provide ./scripts/dist_train.sh that is able to train a model with the specified config file. Run it with:

cd tools/
sh ./scripts/dist_train.sh 1 --cfg_file ./cfgs/m3detr_models/m3detr_waymo_1500.yaml --workers 4

For more options, see ./scripts/dist_train.sh.