I modified the st_gcn_aaai18.py, and want train a new model by using the dataset kinetics and NTU-D, but the accuracy is very low.
So I use the original st_gcn_aaai18.py and train a new model by using the dataset kinetics and NTU-D, the accuracy is very low too. But in theory, its accuracy should as same as the pre-train model, what is wrong with it?
I modified the st_gcn_aaai18.py, and want train a new model by using the dataset kinetics and NTU-D, but the accuracy is very low.
So I use the original st_gcn_aaai18.py and train a new model by using the dataset kinetics and NTU-D, the accuracy is very low too. But in theory, its accuracy should as same as the pre-train model, what is wrong with it?