Skip to content

Latest commit

 

History

History

Datasets

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
COIN-SV
COIN-SV
 
 
CSV
CSV
 
 
DIVING48-SV
DIVING48-SV
 
 
README.md
README.md
 
 
utils.py
utils.py
 
 

README.md

Datasets Configuration Instruction

This project relates to three datasets: COIN-SV, Diving48-SV, CSV. The first two are rearranged based on the existing datasets (COIN, Diving48), and the third is newly collected.

Dataset # Tasks # Videos # Steps # Procedures # Split Videos # Split Pairs
COIN-SV 36 2114 749 37 / 268 / 285 1221 / 451 / 442 21741 / 1000 / 400
Diving48-SV 1 16997 24 20 / 20 / 8 6035 / 7938 / 3024 50000 / 1000 / 400
CSV 14 1941 106 45 / 25 / - 901 / 1039 / - 8551 / 1000 / -

The following are the download instruction and splits division of above datasets.

COIN-SV

Download: We provide a video id list accompained with a script to download videos from YouTube

Splits: train / test / val

NOTE: We truncate each video according to the start and end timestamps provided in the raw annotation.

Diving48-SV

Download: here

Splits: train / test / val

CSV:

Download: BaiduNetDisk (extraction code: 9uyk) / OneDrive

Splits: train / test

Get Data Prepared:

  1. Download videos through the links above.
  2. Transfer videos to frames (resize to 180 $\times$ 320).
  3. Complete the split files with your local data path.

Something else:

  • We also provide video pairs used in our training for reproducing the result reported in the paper if someone interests. (COIN-SV, Diving48-SV, CSV)
  • The labels in COIN-SV and CSV follow the form of A.B where A indicates the task id and B indicates the procedure id in task A. The realistic annotation for labels can be found in label_bank.json in each dataset's folder.