I can't download NTU RGB+D about mmskeleton HOT 11 CLOSED

We don't hold the license to distribute the dataset. You can provide your organization (university, company, or anything else) to the dataset organizers to get the data.

from mmskeleton.

Thank you!
I try it!

from mmskeleton.

Hello, thanks for your fantastic work.
Did you get the dataset? @jackasser
I cannot sent a request even. When I submit, the website shows "500 - Internal server error.
There is a problem with the resource you are looking for, and it cannot be displayed."
Do you have other contact information of the institution? @yjxiong

Many thanks

from mmskeleton.

@GimSungwoo
Me too.
I was displayed 500 - Internal server error.
So, I just download from Kinetics-skeleton.

I want to know other way to train this program.

from mmskeleton.

As I said we don’t hold the permission to distribute the dataset. Please send email to the admnistrators or file an issue at their Github page https://github.com/shahroudy/NTURGB-D/

from mmskeleton.

I got it. Thank so much for your help.

from mmskeleton.

In addition, in order to visualize the output (i.e. when input a video or the json file, the system can output the video which tells us the specific recognized action, plus the skeleton), could you probably give me a suggestion or idea to implement?
Many thanks!

from mmskeleton.

@GimSungwoo
GitHub page https://github.com/shahroudy/NTURGB-D/ gives a good Matlab example about how to visualize a video with skeleton annotations. We will also release the code for generating videos we shown in the README soon.

from mmskeleton.

Cool! Thanks very much. I will try it; We are looking forward to your favorable work as well.

from mmskeleton.

As for Kinetics-skeleton dataset, is it exactly the output (Json file) of the Openpose? or you write a code to change the content. Because I've tried Openpose today, and there is one which put the values of x, y, scole together into one array and it produce one json file for every frame rather than only one whole file for the whole video. What I used is the flag --write_keypoint_json.

from mmskeleton.

@GimSungwoo
Yes. For I/O convenience, we changed the structure and rebuilt database extracted by Openpose.
You can save these data as one whole JSON file or '.npy' file like we did. You can also modify the data feeder to read these disperse JSON files (it will be slow).

from mmskeleton.