brucechou1983/CheXNet-Keras

about linear Dense(1024)

aguang1201 opened this issue · 1 comments

Befor the sigmoid of prediction layer,it's a linear dense of 1024.What is it role?I delete the linear dense and the result is also good.

@aguang1201 You're right. I tried many different output layers, and they actually have similar performance. Finally, I remove the last Dense(1024) layer for CAM.