Official repository of AAAI26 paper: TechCoach: Towards Technical Keypoint-Aware Descriptive Action Coaching
The meta data of EE4D-DescCoach Dataset is constructed in a .json style file:
{
"request-1": {
"clip_st_ed": [
Startint time stamp of the instance,
Ending time stamp of the instance
],
"take_name": "Take name in Ego-Exo4D",
"take_id": Take id in Ego-Exo4D,
"task_name": "Task name in Ego-Exo4D",
"tech_instruction": [
{
"body_part": "SPECIFIC_BODY_PART",
"instruction": "BODY-PART-LEVEL Technical-Point"
],
"raw_comment": [
" ORIGINAL EXPERT COMMENTARY FROM Ego-Exo4D DATASET
],
"response": {
"regular_dict": [
{
"body_part": "SPECIFIC_BODY_PART",
"what_is_done_well": "Technical-Point-level Strength",
"what_can_be_improved": "Technical-Point-level Weakness"
},
...
],
"summarization": "Instance-level Commentary on WHAT_IS_DONE_WELL & WHAT_CAN_BE_IMPROVED."
}
},