Deep residual nets traning

Hello everyone I am working on deep residual nets and I have to train it . Over shape net model. Now I don’t have much idea about traning these big nets.
I have use batch norm,skip connection. In my nets.
But my training loss started decreasing in start and stops after 3-4epochs.
So what may be the reasons. I am using cross entropy with Adam as optimiser. learning rate=.001.