The Moments in Time Recognition Challenge at
CVPR'18 was jointly held with the
ActivityNet Challenge 2018. The goal of this challenge was to identify the event labels depicted
in a 3 second video. The video data came from the Moments in Time dataset, which could be downloaded
here.
The challenge had two tracks:
We will use top-k accuracy on the testing set as the official metrics for this task. For each video, an algorithm will produce $k$ labels $l_{j}$, $j = 1,..,k$. The ground truth label for the video is $g$. The error of the algorithm for that video would be: $$e = \min_{j} d(l_{j}, g),$$ with $d(x,y) = 0$ if $x=y$ and $1$ otherwise. The overall error score for an algorithm is the average error over all videos. We will use $k=1$ and $k=5$ and the winner of the challenge will be selected based on the average of these two errors.
When submitting your results for the challenge, please provide a plain text file containing one video prediction per line. Each line should contain the video filename followed by up to 5 detected class labels, sorted by confidence in descending order:
<filename> <class(1)> <class(2)> <class(3)> <class(4)> <class(5)>