Skip to content

Latest commit

 

History

History
39 lines (32 loc) · 2.2 KB

train_eval.md

File metadata and controls

39 lines (32 loc) · 2.2 KB
Raw

DiffusionDrive Training and Evaluation

1. Cache dataset for faster training and evaluation

# cache dataset for training
python navsim/planning/script/run_dataset_caching.py agent=diffusiondrive_agent experiment_name=training_diffusiondrive_agent train_test_split=navtrain

# cache dataset for evaluation
python navsim/planning/script/run_metric_caching.py train_test_split=navtest cache.cache_path=$NAVSIM_EXP_ROOT/metric_cache

2. Training

If your training machine does not have network access, you should download the pretrained ResNet-34 model from huggingface and upload it to your training machine. You should also download the clustered anchors

Before starting training, ensure that you correctly set the bkb_path to the path of the downloaded pretrained ResNet-34 model. Additionally, set the plan_anchor_path to the path of the downloaded clustered anchors in the file located at /path/to/DiffusionDrive/navsim/agents/diffusiondrive/transfuser_config.py.

python $NAVSIM_DEVKIT_ROOT/navsim/planning/script/run_training.py \
        agent=diffusiondrive_agent \
        experiment_name=training_diffusiondrive_agent  \
        train_test_split=navtrain  \
        split=trainval   \
        trainer.params.max_epochs=100 \
        cache_path="${NAVSIM_EXP_ROOT}/training_cache/" \
        use_cache_without_dataset=True  \
        force_cache_computation=False

3. Evaluation

You can use the following command to evaluate the trained model export CKPT=/path/to/your/checkpoint.pth, for example, you can download the our provided checkpoint from huggingface, and set CKPT=/path/to/downloaded/huggingface_diffusiondrive_agent_ckpt/diffusiondrive_navsim_88p1_PDMS.pth:

python $NAVSIM_DEVKIT_ROOT/navsim/planning/script/run_pdm_score.py \
        train_test_split=navtest \
        agent=diffusiondrive_agent \
        worker=ray_distributed \
        agent.checkpoint_path=$CKPT \
        experiment_name=diffusiondrive_agent_eval