OpenCV dnn not support Squeeze & Excitation Module

asked 2020-04-27 20:31:21 -0600

I got efficientNet model from tensorflow, i can load the pb model successfully ,but it doesn't work when forward, error:-inputs[0][j] == inputs[i][j], i know the problem on the forum, it is the problem of Squeeze & Excitation Module, Does anyone solve it now?

edit retag flag offensive close merge delete

Comments

solved in latest 3.4 / 4.3, please try to update

berak ( 2020-04-28 01:26:14 -0600 )edit

add a comment

1 answer

Sort by » oldest newest most voted

answered 2020-05-01 23:57:03 -0600

cajopencv
1 ●1

i have update to 4.3,but it the problem still exists, i have used optimize_for_inference_lib to optimize the model, but nothing changed,other backbone like resnext50 work great,i use getUnconnectedOutLayersNames to get unconnected Layers info: block4f_se_squeeze/Mean/flatten