Skip to content

Latest commit

 

History

History
62 lines (47 loc) · 3.58 KB

GET_STARTED.md

File metadata and controls

62 lines (47 loc) · 3.58 KB

Getting Started

Train Predefined Models on Standard Datasets

  • This codebase implements distributed training and non-distributed training, which uses MMDistributedDataParallel and MMDataParallel, respectively.

  • All outputs (log files and checkpoints) will be saved to the working directory, which is specified by work_dir in the config file.

  • ⚠️ Important: By default, we evaluate the model on the validation set after each epoch, without using extra tricks including model ensemble and test-time-augmentation (TTA). To ensure fair comparisons, we advise you to follow the same configuration strictly.

  • You can change the evaluation interval by adding the interval argument in the training config.

    train_cfg = dict(type='EpochBasedTrainLoop', val_interval=1)

  • Note: The default learning rate in config files is for 8 GPUs and the exact batch size is marked by the config’s file name, e.g., ‘2xb8’ means 2 samples per GPU using 8 GPUs. According to the Linear Scaling Rule, you might need to set the learning rate proportional to the batch size if you use different GPUs or images per GPU, e.g., lr=0.01 for 4 GPUs * 2 img/gpu and lr=0.08 for 16 GPUs * 4 img/gpu. However, since most of the models in this repo use Adam rather than SGD for optimization, the rule may not hold and users need to tune the learning rate by themselves.

⌛ Train with a single GPU

  • The default command is as follows:

    python tools/train.py ${CONFIG_FILE} [optional arguments]

  • 📝 For example, to train LaserMix with the Cylinder3D backbone on SemanticKITTI under a 10% annotation budget using a single GPU, run the following command:

    python tools/train.py configs/lasermix/lasermix_cy3d_semi_semantickitti_10.py

  • Note: If you want to specify the working directory in the command, you can add an argument --work-dir ${YOUR_WORK_DIR}.

⌛ Train with multiple GPUs

  • The default command is as follows: 
    ./tools/dist_train.sh ${CONFIG_FILE} ${GPU_NUM} [optional arguments]
  • 📝 For example, to train LaserMix with the Cylinder3D backbone on SemanticKITTI under a 10% annotation budget using four GPUs, run the following command: 
    ./tools/dist_train.sh configs/lasermix/lasermix_cy3d_semi_semantickitti_10.py 4
  • Note: Optional argument: --cfg-options 'Key=value', which overrides some settings in the used config.

⌛ Train with multiple machines

  • If you run MMDetection3D on a cluster managed with slurm, you can use the script slurm_train.sh. This script also supports single-machine training.

    [GPUS=${GPUS}] ./tools/slurm_train.sh ${PARTITION} ${JOB_NAME} ${CONFIG_FILE} ${WORK_DIR}

  • 📝 For example, to train LaserMix with the Cylinder3D backbone on SemanticKITTI under a 10% annotation budget using slurm, run the following command:

    GPUS=4 ./tools/slurm_train.sh partition lasermix configs/lasermix/lasermix_cy3d_semi_semantickitti_10.py /logs/semi/semantickitti/lasermix_cy3d_10

Test Existing Models on Standard Datasets

Coming soon.