First of all, I searched related issues in FAQ, blogs, etc., but I didn't understand the tutorial on how to build my own model. Maybe the way I searched for the answer was wrong. Looking forward to your reply.
I read tutorials on how to write config files etc but didn't find a way to solve my problem, Like use R(2+1)D conv in Slowfast or STN.
According to the method of configs/recognition/r2plus1d/xxx.py, I tried to add the setting dictionary conv_cfg=dict(type='Conv2plus1d') to the configuration file, but the program cannot be executed and an error is reported.
I want to know if the above idea of building my own model is correct, and I hope to get more tutorials to build my own model. Can different backone, common and heads in mmaction/models be combined at will?
Thanks !!
First of all, I searched related issues in FAQ, blogs, etc., but I didn't understand the tutorial on how to build my own model. Maybe the way I searched for the answer was wrong. Looking forward to your reply. I read tutorials on how to write config files etc but didn't find a way to solve my problem, Like use R(2+1)D conv in Slowfast or STN. According to the method of configs/recognition/r2plus1d/xxx.py, I tried to add the setting dictionary conv_cfg=dict(type='Conv2plus1d') to the configuration file, but the program cannot be executed and an error is reported. I want to know if the above idea of building my own model is correct, and I hope to get more tutorials to build my own model. Can different backone, common and heads in mmaction/models be combined at will? Thanks !!