search

motional / nuplan-devkit

The devkit of the nuPlan dataset.
https://www.nuplan.org
Other
673 stars 129 forks source link

Does the caching process the scenarios? #259

Closed yaqlee closed 1 year ago

yaqlee commented 1 year ago

when I run caching with part of the full data, it takes too long. The log showed as this:

INFO {/root/code/nuplan-devkit/nuplan/planning/training/experiments/caching.py:148} Starting dataset caching of 6075752 files...

According to Fact Sheet - nuPlan Dataset v1.0 in https://www.nuscenes.org/nuplan, the scenarios counts 3 millions at most. Why are there 6075752 files to preprocess? I guess the count was magnitude of lidarpcs files, right? My scripts was like this:

cfg = hydra.compose(config_name=CONFIG_NAME, overrides=[ f'group={str(SAVE_DIR)}', f'cache.cache_path={str(SAVE_DIR)}/cache', 'py_func=cache', 'cache.force_feature_computation=True', 'distribute_by_scenario=True', '+training=training_vector_model', #

raster model that consumes ego, agents and map raster layers and regresses the ego's trajectory

    'scenario_builder=nuplan',  # use nuplan trainval database
    'scenario_filter.limit_total_scenarios=0.50',  # Choose 500 scenarios to train with, int; float should be in (0,1)
])

Did the caching run correctly?