Evaluation of ego vehicle trajectory prediction


I have made a model which uses the past 2 key frames as input data (front_cam, translation and position) and then predicts the 12 future positions and orientations for the ego vehicle only (6 seconds). I want to evaluate my model with the data from the NuScenes prediction challenge (https://www.nuscenes.org/prediction?externalData=all&mapData=all&modalities=Any). I obtain the test set with: test_dataset = get_prediction_challenge_split(“train_val”, dataroot=dataset_path).

  1. Do I have to predict trajectories for all instances in the sample?
  2. How do I get the input data my model needs for predicting future trajectory?