Hi, I am following this procedure:
- Train a network with
train.py, and save the model with
- Next I load the model in
However, now I notice the model gives entirely different results if I call
.eval() or not. The model includes a couple of BatchNorm2d and Dropout layers, and performs fine without calling
.eval(), but I was hoping for a slight improvement in performance with
.eval() activated. What are thinks to look out for when encountering this problem?