Open BrownOfSummer opened 5 years ago
Using three models can also deployed simply in C++.
How to convert the three models of PNet, RNet, Onet into single pb file, so it simple to deploy in C++?
I think it will be little complex to convert three model to one pb file. A pb file looks like only contain tensor, tensor'op and their paramter.
MTCNN needs run NMS after run PNet and RNet. But in this project looks like NMS code with numpy,not tensorflow, the PRO Net connect part code with python as well..
therefore,maybe you need define your own tensor NMS and use tensor connect three model.
Using three models can also deployed simply in C++.
How to convert the three models of PNet, RNet, Onet into single pb file, so it simple to deploy in C++?
I think it will be little complex to convert three model to one pb file. A pb file looks like only contain tensor, tensor'op and their paramter.
MTCNN needs run NMS after run PNet and RNet. But in this project looks like NMS code with numpy,not tensorflow, the PRO Net connect part code with python as well..
therefore,maybe you need define your own tensor NMS and use tensor connect three model.
Tensorflow can do NMS, the difficult is how to do image pyramid
How to convert the three models of PNet, RNet, Onet into single pb file, so it simple to deploy in C++?