/CharadesDet

Charades Object Detection Dataset (ICCV 2017)

Primary LanguageMATLAB

Charades Object Detection Dataset (ICCV 2017)

Citation

If you use our dataset in your research, please use the following BibTeX entry.

@inproceedings{YuanyuanICCV2017,
    Author = {Yuan Yuan and Xiaodan Liang and Xiaolong Wang and Dit-Yan Yeung and Abhinav Gupta},
    Title = {Temporal Dynamic Graph LSTM for Action-driven Video Object Detection},
    Booktitle = {ICCV},
    Year = {2017},
}

We label all the videos in the Charades Validation video set, which can be downloaded from here.

The annotation is Annotations.tar.gz in this repo. There are 1812 videos labeled in total and 8 boxes per frame in average.

There are 35 classes 'charade_object_classes.txt'.

We label one frame every second, and we extract the frames in the original frame rate of the videos with:

extract_frames.m

Note that the frame rate for every video is different, You will need to extract the frames with this script. We did NOT use the 24fps jpg images from the official webpage.