Skip to main content

Proficiency Estimation

Demonstrator and demonstration proficiency are available in a single JSON file per split containing a list of annotations for a given take.

Demonstration Annotations

For each take, you will find:

  • take_uid: unique id given to this take
  • good_executions: list of annotations that were deemed to be a good execution by the expert
    • video_time: the single point in time of the annotation in seconds
    • list: the commentary provided by the expert for the particular annotation
    • expert_uid: a unique identifier for the annotator who provided the annotation
    • video_name: the video name associated with the take
  • tips_for_improvement: list of annotations that were deemed to be suboptimal with a tip for improvement
    • video_time: the single point in time of the annotation in seconds
    • list: the tip provided by the expert for the particular annotation
    • expert_uid: a unique identifier for the annotator who provided the annotation
    • video_name: the video name associated with the take

Additional take fields (also available in the global metadata):

  • task_id: unique id for the task type
  • scenario_name: name of the high level activity happening in this take
  • task_name: name of the particular task within the scenario
  • university_id: id of the university where the take was recorded
  • origin_participant_id: participant id unique to the university

Sample JSON

Here is a JSON snippet showing a sample annotation:

[
{
"task_id": 6002,
"task_name": "Basketball Drills - Reverse Layup",
"scenario_name": "Basketball",
"university_id": "ego4d-unc",
"origin_participant_id": "30",
"take_uid": "0021a226-05b9-4ca4-9e2a-cc4fae3357f3",
"video_paths": {
"ego": "takes/unc_basketball_03-30-23_02_34/frame_aligned_videos/aria02_214-1.mp4",
"exo1": "takes/unc_basketball_03-30-23_02_34/frame_aligned_videos/cam01.mp4",
"exo2": "takes/unc_basketball_03-30-23_02_34/frame_aligned_videos/cam02.mp4",
"exo3": "takes/unc_basketball_03-30-23_02_34/frame_aligned_videos/cam03.mp4",
"exo4": "takes/unc_basketball_03-30-23_02_34/frame_aligned_videos/cam04.mp4"
},
"good_executions": [
{
"video_time": 0.983072,
"list": [
"Getting lowered towards the ground",
"Bending his legs in the athletic stance",
"Getting towards the right side of the rim quicker with more balance and power"
],
"expert_uid": "6jm95g",
"video_name": "unc_basketball_03-30-23_02_34",
},
...
],
"tips_for_improvement": [
{
"video_time": 87.720592,
"list": [
"It\'s taking him, giving him the momentum to move his body behind.",
"It\'s going to follow with the left foot and up."
],
"expert_uid": "2g7km5",
"video_name": "unc_basketball_03-30-23_02_34",
},
...
],
},
...
]

Demonstrator Annotations

For each take, you will find:

  • take_uid: unique id given to this take
  • proficiency_score: the skill estimation of the participant - possible values: [Novice, Early Expert, Intermediate Expert, Late Expert]

Additional take fields (also available in the global metadata):

  • task_id: unique id for the task type
  • scenario_name: name of the high level activity happening in this take
  • task_name: name of the particular task within the scenario
  • university_id: id of the university where the take was recorded
  • origin_participant_id: participant id unique to the university

Sample JSON

Here is a JSON snippet showing a sample annotation:

[
{
"task_id": 7018,
"task_name": "LosAndes_V2_R1",
"scenario_name": "Rock Climbing",
"university_id": "ego4d-university-sa",
"origin_participant_id": "133",
"take_uid": "000a19fe-776e-4c88-b0c3-2fad016a6025",
"video_paths": {
"ego": "takes/uniandes_bouldering_028_7/frame_aligned_videos/aria01_214-1.mp4",
"exo1": "takes/uniandes_bouldering_028_7/frame_aligned_videos/cam01.mp4",
"exo2": "takes/uniandes_bouldering_028_7/frame_aligned_videos/cam02.mp4",
"exo3": "takes/uniandes_bouldering_028_7/frame_aligned_videos/cam03.mp4",
"exo4": "takes/uniandes_bouldering_028_7/frame_aligned_videos/cam04.mp4"
},
"proficiency_score": "Early Expert"
},
...
]