Skip to content

DATASET.md

Latest commit

 

History

History
36 lines (23 loc) · 1.85 KB

DATASET.md

File metadata and controls

36 lines (23 loc) · 1.85 KB

Dataset Preparation

We provide our labels in data_list.

Kinetics

The Kinetics Dataset could be downloaded via the code released by ActivityNet:

  1. Download the videos via the official scripts. You also can download videos as show in cvdfoundation.

  2. After all the videos were downloaded, resize the video to the short edge size of 320, then prepare the csv files for training, validation, and testing set as train.csv, val.csv, test.csv in data_list/k400. The format of the csv file is:

path_to_video_1,label_1
path_to_video_2,label_2
path_to_video_3,label_3
...
path_to_video_N,label_N

Note that we use decord to decode the Kinetics videos on the fly.

Since some videos may no longer be available, it will lead to small performance gap. If necessary, we will provide our version of Kinetics-400.

Something-Something V2

  1. Please download the dataset and annotations from dataset provider.

  2. Download the frame list from the following links: (train, val).

  3. Extract the frames at 30 FPS. (We used ffmpeg-4.1.3 with command ffmpeg -i "${video}" -r 30 -q:v 1 "${out_name}" in experiments.) Please put the frames in a structure consistent with the frame lists.

Please put all annotation json files and the frame lists in the same folder, and set DATA.PATH_TO_DATA_DIR to this folder path e.g., data_list/sthv1 and data_list/sthv2. Set DATA.PATH_PREFIX to be the path to the folder containing extracted frames.

Since the web page of 20BN is not accessible, we will provide the raw datasets if necessary.