code for Searching Actions on the Hyperbole.

The current code includes all the essentials to re-implement our paper,
The demo code is a raw version, please leave comments if there is any difficuty to use it.

Requirements

pytorch == 1.4.0
geoopt == 0.2.0

If you fail to use this code, please email me (uestc.longteng@gmail.com)

Data Hierarchy and Seen/Unseen Split.

activity_net_depth.csv
kinetics_depth.csv
moments_depth.csv

Data files

I have uploaded the extracted video features (~8GB) to Microsoft Onedrive, please contact me for data sharing.

Pretrained 300-D action embedding:

cone_activity_net300.pth
cone_kinetics_300.pth
cone_moments_300.pth

Reference

Searching for Actions on the Hyperbole

Teng Long, Pascal Mettes, Heng Tao Shen, Cees G. M. Snoek

@InProceedings{Long_2020_CVPR,
author = {Long, Teng and Mettes, Pascal and Shen, Heng Tao and Snoek, Cees G. M.},
title = {Searching for Actions on the Hyperbole},
booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}
}