gzoumpourlis/FewShotKinetics_csv_lists

Parsing the dataset of "Compound Memory Networks for Few-shot Video Classification"

The data splits of the dataset (let's call it "Kinetics-100" for convenience) that was introduced by the publication "Compound Memory Networks for Few-shot Video Classification - ECCV 2018", in text format, as given by the authors in their github repo, can be found in data_splits folder. (source : https://github.com/ffmpbgrnn/CMN ) These data splits are re-written in .csv format (the same format with that of the data splits of the full Kinetics-400 dataset), in data_splits_100 folder, by executing parse_classes.py.

The data splits of the Kinetics-400 dataset, can be found in .csv format, in data_splits_400 folder. ( source : https://github.com/activitynet/ActivityNet/tree/master/Crawler/Kinetics/data )

To download the video files of the dataset splits of Kinetics-100, you'll need the Kinetics crawler, and the splits in .csv format (as mentioned above, you can find the splits in the data_splits_100 folder) :

1. Clone the Kinetics dataset crawler: git clone https://github.com/activitynet/ActivityNet/tree/master/Crawler/Kinetics

2. Copy the .csv files from the data_splits_100 folder, to the data folder of the crawler's repo

3. Copy the download_Kinetics_100.py file to the root directory of the crawler's repo, i.e. in the same directory with the crawler's download.py file

4. Create the folder where the Kinetics-100 will be downloaded and stored. mkdir dataset

5. Run the script to download the train/val/test splits of Kinetics-100 python download_Kinetics_100.py