I created the same issue here, but I will share here as well.
According to the submission guidelines of future hand prediction challenge, the submission format should be a .json file with following format:
{"ClipId1_FrameId1":
[p3_left_hand_x, p3_left_hand_y, p3_right_hand_x, p3_right_hand_y,
p2_left_hand_x, p2_left_hand_y, p2_right_hand_x, p2_right_hand_y,
p1_left_hand_x, p1_left_hand_y, p1_right_hand_x, p1_right_hand_y,
p_left_hand_x, p_left_hand_y, p_right_hand_x, p_right_hand_y,
c_left_hand_x, c_left_hand_y, c_right_hand_x, c_right_hand_y],
"ClipId1_FrameId2":
[p3_left_hand_x, p3_left_hand_y, p3_right_hand_x, p3_right_hand_y,
p2_left_hand_x, p2_left_hand_y, p2_right_hand_x, p2_right_hand_y,
p1_left_hand_x, p1_left_hand_y, p1_right_hand_x, p1_right_hand_y,
p_left_hand_x, p_left_hand_y, p_right_hand_x, p_right_hand_y,
c_left_hand_x, c_left_hand_y, c_right_hand_x, c_right_hand_y],
...
}
However, we only know clip_id, clip_uid, and, video_uid for test data:
{
"version": "1",
"date": "220217",
"description": "FHO Hands & Objects",
"manifest": "s3://ego4d-consortium-sharing/public/v1/full_scale/manifest.csv",
"split": "test",
"clips": [
{
"clip_id": 3,
"clip_uid": "de4a85e5-1809-4547-ae11-9161aa9fdbe9",
"video_uid": "31ea56e6-7a99-4b91-b74b-3b1e6de05e99"
},
{
"clip_id": 2056,
"clip_uid": "83acc738-4435-451d-a38f-2a3a6cd85879",
"video_uid": "4ce9f93b-f44c-46fe-9050-3452d2f39abb"
},
{
"clip_id": 10,
"clip_uid": "bd5e0084-7bf1-4e34-abba-cc3f6a370213",
"video_uid": "cde41c4f-50d1-4910-9f2a-4c7b6987df92"
},
...
}
Then, I would like to know how can we know which frame is corresponding to the last observable frame (one frame before pre45 frame)?