Skip to content

Latest commit

 

History

History
35 lines (25 loc) · 1.23 KB

DATASET.md

File metadata and controls

35 lines (25 loc) · 1.23 KB

Dataset Preparation

Mini-Kinetics-JPG

  • Download videos
    • Locate test set in video_directory/test.
  • Convert from avi to jpg files using dataset_utils/generate_video_jpgs.py
python dataset_utils/generate_video_jpgs.py mp4_video_dir_path jpg_video_dir_path kinetics --n_jobs NUM_CPU_CORES
  • Generate annotation file in json format similar to ActivityNet using dataset_utils/kinetics_json_parallel.py
    • The CSV files (kinetics-600_{train, val, test}.csv) are included in the crawler.
python dataset_utils/kinetics_json.py csv_dir_path 600 jpg_video_dir_path jpg dst_json_path --n_jobs NUM_CPU_CORES

Kinetics-JPG

  • Download videos
    • Locate test set in video_directory/test.
  • Convert from avi to jpg files using dataset_utils/generate_video_jpgs.py
python dataset_utils/generate_video_jpgs.py mp4_video_dir_path jpg_video_dir_path --vid_ext avi --size 452 --n_jobs NUM_CPU_CORES
  • Generate annotation file in json format similar to ActivityNet using dataset_utils/kinetics_json.py
    • The CSV files (kinetics-400_{train, val, test}.csv) are included in the crawler.
python dataset_utils/kinetics_json.py csv_dir_path 400 jpg_video_dir_path jpg dst_json_file