[project page] [Torch code for T-CAM model]
Tools and datasets collected for the CVPR 18 paper
D. Wei, J. Lim, A. Zisserman, W. Freeman. "Learning and Using the Arrow of Time." in CVPR 2018.
- data_ReverseFilm
- data_MJPEG-AoT
- data_TA180
- tool_crawl
- [] data_Flickr-AoT
- [] data_Kinetics-AoT
- [] tool_process
We select clips whose camera motion can be well-stabilized.
Dataset | #Clips | Description |
---|---|---|
MJPEG-AoT (new) | 16.9k | overlapped with the Vimeo-90K dataset |
ReverseFilm (new) | 67 | movie clips using the reverse motion filming technique |
TA180 | 165 | subset of the previous TA180 dataset |
Kinetics-AoT | 57k | subset of the Kinetics dataset |
Flickr-AoT | 147k | subset of the Flickr-100M + Extra Flickr dataset |
Name | Usage |
---|---|
crawl | crawling Vimeo urls by query and user names |
process | scene detection and camera stabilization |
util | utility function |
Please cite our paper if you find it useful for your work:
@inproceedings{wei2018learning,
title={Learning and Using the Arrow of Time},
author={Wei, Donglai and Lim, Joseph J and Zisserman, Andrew and Freeman, William T},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages={8052--8060},
year={2018}
}