results/episode_task_suite/object_relevance/metrics.json

{
  "micro_f1": 0.18034382095361662,
  "macro_f1": 0.06329638076675959,
  "exact_match": 0.005747126436781609,
  "precision": 0.16106604866743918,
  "recall": 0.20486366985998525,
  "task": "object_relevance",
  "input": "all non-caption modalities -> current relevant object set",
  "split": "chronological",
  "num_windows": 1161,
  "num_train_windows": 813,
  "num_test_windows": 348,
  "num_objects": 34,
  "task_display_name": "Object Relevance Prediction"
}