THUMOS 2014+2015 Action Recognition Dataset

THUMOS14 is a large-scale video dataset, including 1,010 validation videos and 1,574 test videos from 20 categories. Among all the videos, the validation set and test set have 220 and 212 videos with time annotations, respectively.

The latest version of the dataset is THUMOS 2015, which contains more than 430 hours of video data and 45 million frames (70% more than THUMOS14), including the following subsets:

• Training set: 13,000+ time-resized videos from 101 action classes.
• Validation set: over 2,100 videos of unscaled duration with temporal annotations of actions.
• Background set: approximately 3,000 relevant videos that are guaranteed not to contain any instances of the 101 actions.
• Test set: over 5,600 unscaled videos with implicit ground truth.

This dataset includes two versions: 2014 and 2015.

Dataset decompression password:THUMOS14_REGISTERED