This is a official about our casr dataset, the paper link will be added soon. Dataset link: http://adas.cvc.uab.es/casr/ Video folder have 89 videos. casr_annotation.pickle is the annotation of 4 cyclists. youtube_annotation.pickle is the annotation of Youtube videos. Here is the usage of this dataset. mkdir data unzip annotations.zip -d data cd data mkdir videos Then you need to download the videos and put it into the folder of videos cd .. You need to run 'sh extract_images.sh' for generating the images from videos. It need several hundred gigabit space. run the script of view.py to extact the sequeces of cyclists. For example: 'python2 view.py --dataset casr --track_number 0', see the view.py for more information. Please cite our IEEE journal paper if you use our dataset and code. @article{zhijie2019intention, title={Intention Recognition of Pedestrians and Cyclists by 2D Pose Estimation}, author={Zhijie Fang, Antonio M. López}, journal={IEEE Transactions on Intelligent Transportation Systems}, year={2019} } You can also download from the arxiv: @article{zhijie2019intention, title={Intention Recognition of Pedestrians and Cyclists by 2D Pose Estimation}, author={Zhijie Fang, Antonio M. López}, journal={arXiv:1910.03858}, year={2019} }