detection2d

tools/detection2d

Training and evaluation tools for 2D Detection.

• Support priority: Tier A
• Supported dataset
  • COCO dataset
  • T4dataset
• Other supported feature
  • Add unit test

1. Set environment

• See setting environment

2. Prepare dataset

2.1. COCO dataset

TBD

2.2. T4dataset

• (Choice) For traffic light recognition of fine detector

python3 tools/detection2d/create_data_t4dataset.py --config autoware_ml/configs/detection2d/dataset/t4dataset/tlr_finedetector.py --root_path ./data/tlr/ --data_name tlr -o ./data/tlr_pedcar

3. Train

3.1. Change config

• If you need, change parameters for your experiment
• You can change experiment name by file name

3.2. Train

• You can use docker command for training as below.
  • See each projects for detail command of training and evaluation.

python tools/detection2d/train.py {config_file}

• You can use docker command for training as below.

docker run -it --rm --gpus '"device=1"' --name autoware-ml --shm-size=64g -d -v $PWD/:/workspace -v $PWD/data:/workspace/data autoware-ml bash -c '<command for each projects>'

4. Analyze

4.1. Evaluation

• Evaluation

python tools/detection2d/test.py {config_file} {checkpoint_file}

4.2. Visualization

python tools/detection2d/image_demo.py {image path} {config_file} --weights {pth_file}

5. Deploy

See each projects