sy-zhang/geometric-feature-release

Codes for our WACV2017 paper: "On Geometric Features for Skeleton-Based Action Recognition Using Multilayer LSTM Networks"

Hi there 👋

geometric-feature-release's Issues

Sample Skeleton File?

In hdf5 file, there are many feature values for videos and class respectively.
And dimension are various. For example, dimension of walk s10_e02_a01 for JL_D is (39,612) and clapHands is (30,612). What is difference between them? When I check their frame count, for walk has 124 frames and for clapHands has 144. Moreover, why you choose 612 dimension for JL_D and 512 dimension for others, etc?

more detail of this project

would you mind sharing more code and even the pretrained model? After investigating, I feel your work is good, and I am interested in it and want to try it. But, there are some confused with the details

Could you provide me script of LSTM network with Geometric-feature?

Hello.

I read your paper. Your work is excellent.
I found geometric feature scripts.
But, I have no idea how to implement LSTM network using these features as input. And how long take time for training? Moreover, for 16 joints, lines are 120 and planes are 560. But, for 120 joints, how to calculate for lines and planes (in table 1).
So, if you do not mind, could you provide me sample scripts of this network with fusion parts.