kennymckormick
I have a quick question!
Thanks for your awesome works! I have a quick question. I read a paper relate to DG-STGCN. I'd like to make a system using pyskl and openpose. specifically, if there are some videoes, I infer the video using openpose model. and then i will get the skeleton sequence. DG-STGCN take that sequence as input. and then it infers action category. it is the system which I will make.
here is the question.
To infer action category, i have to get DG-STGCN model's weight file. if I get it, I can use various datasets. but they are all different skeleton sequence data. it means they are used different prescribe graphical structures. If i use NTURGB+D dataset to train, does my system work on well? ( NTURGB+D : 3D skeleton sequence, openpose : 2D skeleton sequence )
thanks for reply. and then I had another question. I would like to use different lose function. How could i?
thanks for reply. and then I had another question. I would like to use different lose function. How could i?
pyskl文件夹里可能找到相关代码
Currently, we have not supported GCN models + Kinetics, we will support them shortly.
I have one more question. can I export stgcn++ model to onnx?
Did you solve it? st-gcn++
@Wooho-Moon how to exchange stgcn++ model to onnx?
I exchange stgcn++ model to onnx. u wanna export, u need to exchange some operators. I use onnx opset version 11. so i can't use the operator enisum. so u have to replace that operator right ways. I thought there were other exchanges in there. But, I don't think excatly what they are.
