TJU Dataset


Human action analysis has been an active research topic in recent years. We contribute a public dataset with 1936 action sequences (22 actions * 22 subjects * 4 time). The initial goal of preparing for it is for evaluating the local space-time descriptor-based methods for human action recognition. Hopefully, it will be widely utilized for the evaluation of other methods.


1) There are limited action types in Weizmann, KTH, and MSR Action Dataset I&II. The low resolution of videos in Weizmann and KTH has negative influence on local salient point extraction.

2) Action3D does not contain RGB data and cannot be used for our work.

3) Most actions in MSRDailyActivity3D involve the humans-object interactions, like read book, call cellphone et al. It is difficult to extract local salient points for these actions since there is no salient motion. Therefore, these kinds of action are not proper for evaluating the local space time descriptor-based methods.

4) DHA has 23 kinds of actions but only have 21 samples for each action. Therefore it is not suited for learning-based methods.

Data Description

Each of 22 actions was performed twice by 22 subjects (2 more subjects compared to the reported subject number). There are totally 1936 action sequences. All sequences were taken with a static camera with 20fps frame rate and 640×480 resolution.

All the action samples are in the “data”folder and named with the action index, subject index, and repeating index. For example, “a01_p01_t01”denots the first action performed by the first person for the first time.

Train/Test Splits
Training: person 1-8
Validation: person 9-14
Test: person 15-22

Action Category

For action type selection, we inherited the popular action types in the 6 popular datasets, Weizmann, KTH, MSRDailyActivity3D, Action3D, MSR Action Dataset I&II and DHA. We kept 22 actions to cover most of the popular types with salient motion in upper body and lower body.

Actions in this data set include:
1.Boxing; 2.Side boxing; 3.One hand wave; 4.Two hands wave; 5.Hand clap; 6.Side bend; 7.Forward bend; 8.Draw X; 9.Draw tick; 10.Draw circle; 11.Tennis serve; 12.Tennis swing; 13.Walking; 14.Side walking; 15.Jogging; 16.Running; 17.Jacks; 18.Jump; 19.Jump in place; 20.Forward kick; 21.Side kick; 22.Sit down.


We build one ftp server for downloading this dataset. The participants also need to download and file Agreement and Disclaimer Form and send it back to with your register email. We will then email you the instructions to download the dataset.

Note: The participants also need to download and file Agreement and Disclaimer Form and send it back to us with your register email.

RGB Images