The Kinetics Human Action Video Data Set [1,2] is a data set of about 500,000 YouTube video clips that depict persons engaging in physical actions.There are two types of videos: one that encompasses videos depicting one of 400 unique actions (Kinetics-400), and the other that features videos depicting one of 600 actions (Kinetics-600). These actions can vary from simple motions like applauding, to something that is much more complex such as wrapping a present or weaving a basket.
For this contest, you will be asked to work with dataset Kinetics 400.
You are expected to develop a machine learning algorithm that can, after being trained on a training set derived from Kinetics 400 [1], determine what action is being depicted in a given test video. Training, validation, and test sets are given in [1].
The challenges you will likely encounter are as follows, and will better explain why this development is both so important and so valuable:
You should follow the submission guidelines of Kaggle as given at the link below to submit your complete solution:
https://www.kaggle.com/WinningModelDocumentationGuidelines
All entries of the competition that are received by (insert datetime) will be judged on a rubric that will not be disclosed to competitors. After evaluation by an expert committee, we will select the winner for each of the following categories:
Below is a list of the important dates of the competition, which are hard deadlines. Please make sure you submit your team’s final algorithm and associated files as an organized .zip archive by the deadline date. We do not consider late submissions.
Final submission date: July 31, 2019 11:59 PM
Winning team will be announced on: August 31, 2019
If you have any questions related to the competition, you can contact
| {sharad,clint}@iitgoa.ac.in |
