NTU DB Data Loader

• Please refer to the link below for more detailed information about NTU DB: (http://rose1.ntu.edu.sg/datasets/actionrecognition.asp) (https://github.com/shahroudy/NTURGB-D)

• This python code provides a NTU DB data loader which loads the data from the hard disk and make them a structured object. The folder structure of your NTU DB should be organized as described in the following section, but if you want to use your own folder, you can modify the part of the setting path of this code and use it.

• We also provide a function for marking exception file. There are some errors in data which the skeleton and the performer are not matched or missing. These data can be excluded by marking them and putting on a list. After making the exception file list, you can utilize it to separate the error data from the original.

Update 2019/04/30

Update Note:

1. Append mode was removed. Now the appending is determined by show mode. If the show mode is 0(show all) or 1(show only filtered files), will be merged automatically.
2. Re-inclusion mode was added. You might make mistakes that add a normal file to the exception list while filtering the data. In that situation, you can remove the normal file name from the exception list by pressing 'z' button in show mode 2(show only error data).
3. A bug was fixed. If a skeleton was not found in a video, then the video is added to the exception list.

Update 2019/03/11

Update Note:

1. Show mode. If the "show_mode" variable is set to 0, shows all the original NTU DB data. If 1, shows only normal data with the exception of the files listed on the exception file. And if 2, loads and shows only the error files from exception file list.
2. Append mode. If you already have the exception file, and you missed excluding some error data from original data, you can append the error data to the list that you missed before. By setting the "with_append" and "save_exception_list" variables "True", you can do the appending.
3. Skeleton data class was moved to the independent file " Skeleton.py". Therefore you should import the classes to use the skeleton class.

Data Loading

***Notice

Before you use this code, Your NTU database files should be organized as following.

Example)
/*your own path */NTU_DB/Daily_Actions/A022
/*your own path */NTU_DB/Mutual_Conditions/A050 ...

The pairs of video clip and skeleton file corresponding to each action class should be in each subfolder.

Example)
In /*your own path */NTU_DB/Daily_Actions/A022, it contains.. S001C001P001R001A022_rgb.avi, S001C001P001R001A022.skeleton, ...

After data loading, the skeleton data information will be transformed to the python object which consisting of 3D location, 2D location of RGB/Depth/IR, orientation of each joint and etc. for every frame and every body.

Drawing Skeleton

In each frame, the skeleton data corresponding to the performer's body is drawn with each joint and the lines connecting the joints. This code only provides skeleton drawing for RGB, not for Depth or IR yet.

Quick View

You can quickly search the video and skeleton data at once by quick view function. While you are watching the video, press the button 'e' to move next video immediately, press the 'q' button to move previous video. 'ESC' button can make you close the program.

Marking Error Data

If you found the error data which the unmatched or missing skeleton in the process of checking videos,
you can simply mark the current video and skeleton by pressing 'x' button. Once you finish the checking process by reaching the end of the class's file or 'ESC', then the name list of the marked data is generated and saved as a text file after deduplication.