LOVEU
(LOng-form VidEo Understanding)
Workshop & International Challenge @ CVPR'21 June 25
Invited Speakers:
Ivan Laptev
INRIA Paris
Andrew Zisserman
University of Oxford & DeepMind
Chao-Yuan Wu
Facebook AI
Raquel Urtasun
Waabi, University of Toronto
Organizers:
Facebook AI, NUS
Stan LEI
NUS
Linchao Zhu
UTS
Xiaohan Wang
UTS
Karttikeya Mangalam
UC Berkeley
Weiyao Wang
Facebook AI
Yi Yang
UTS
Lorenzo Torresani
Facebook AI & Dartmouth
Kristen Grauman
UT Austin
Matt Feiszli
Facebook AI
Jitendra Malik
UC Berkeley
News:
Jun 30, 2021: recording videos, slides, reports have been uploaded and the links will be posted in Program. Thank you all very much for your interests!
June 28, 2021: Congratulations to Team Yonsei_CIPLAB (First Place Award), Team ByteVideo (Second Place Award), Team Visual Analysis of Humans (Third Place Award) on winning Track 1, and Team SAIC-KAUST (Winning Award) on winning Track 2 in LOVEU Challenge! Thanks to all the participants!
May 27, 2021: we made leaderboard for Track 1 public since participants have no access to their latest submission score after leaderboard being hided on CodaLab.
May 26, 2021: we extend the registration&submission deadline to Jun 08, 2021 (11:59PM Pacific Time) for Track 2. Please refer to Track 2 for more details.
May 26, 2021: note that if you participate in both Track 1 and Track 2, duplicate submissions or submissions with significant content overlap are NOT allowed.
May 21, 2021: we update the portal for Track 1 Report Submission, Track 2 Registration and Track 2 Report Submission.
For any generic questions, please post on the forum: https://competitions.codalab.org/forums/27005/; for more personal inquires, please email Xiaohan & Stan.
April 29, 2021: we slightly updated the video list of Kinetics-GEBD. As a summary:
The updated version removes 943 Test Set videos that we found to be overlapped/duplicated with our Train Set. Note this is just a small portion of our Test Set (5.5%).
For some videos like ‘NQcK4qPOqtg’ - its total duration is 207s long but the trimmed clip position provided by the original Kinetics dataset is from 232s to 242s. We remove all such problematic Kinetics videos (4 in Train, 2 in Val and 2 in Test).
Finally note that it is alright if you include the detection results of these videos in your submission file since we will ignore them for evaluation. But you may want to remove these videos too in order to save computation.
April 5, 2021: given popular requests, we plan to split track 1 into two scenarios/sub-tracks:
no constraint of additional supervision for training upstream models and additional training video data (our previous track 1 setting)
cannot use additional supervision for training upstream models or additional training video data (so that is friendly to the team of limited compute resource)
More details have been updated in Track 1 website page and whitepaper. If you have any suggestion or feedback, please feel free to contact us. We will consider them and will finalize all details by April 10 (although we do not expect any major change).
Mar. 2021: Annotations & competition details released.
Dec. 2020: Website launched.