New THOTH website available here

Spatio-temporal object detection proposals

Dan Oneata    Jerome Revaud    Jakob Verbeek    Cordelia Schmid



You can find the code here.


We have re-annotated the UCF Sport dataset every 20 frames. You can find our annotations here. The archive contains a text file for the 150 videos in the dataset. Each line of the annotation files has the following format: frame_num x_pos y_pos width height. The mappings between the video IDs and the video names are in the videos.txt file, but we have used the train-test split defined by the files train.txt and test.txt.


    AUTHOR = {Oneata, Dan and Revaud, Jerome and Verbeek, Jakob and Schmid, Cordelia},
    TITLE = {Spatio-Temporal Object Detection Proposals},
    BOOKTITLE = {{ECCV 2014 - European Conference on Computer Vision}},
    PUBLISHER = {Springer},
    YEAR = {2014},
    MONTH = Sep,
    ADDRESS = {Zurich, Switzerland},
    URL = {},


ECCV 2014

Spatio-temporal object detection proposals
Dan Oneata, Jerome Revaud, Jakob Verbeek and Cordelia Schmid,
ECCV'14, September, 2014.   pdf