any wise way to implement a residual network with 1000 layers ?

[wagamamaz]

No description provided.

[zsdonghao]

@wagamamaz you can use for loop to define it.

lrelu = lambda x: tl.act.lrelu(x, 0.2)
...
net_h1 = Conv2d(net_h0, df_dim*2, (4, 4), (2, 2), act=None,             
        padding='SAME', W_init=w_init, b_init=b_init, name='e_h1/conv2d')
net_h1 = BatchNormLayer(net_h1, #act=lrelu, 
       is_train=is_train, gamma_init=gamma_init, name='e_h1/batchnorm')

for i in range(1000):
      net = Conv2d(net_h1, df_dim*1, (1, 1), (1, 1),
          padding='SAME', W_init=w_init, b_init=b_init, name='e{}/c1'.format(i))
      net = BatchNormLayer(net, act=lrelu,
          is_train=is_train, gamma_init=gamma_init, name='e{}/b1'.format(i))
      net = Conv2d(net, df_dim*1, (3, 3), (1, 1),
          padding='SAME', W_init=w_init, b_init=b_init, name='e{}/c2'.format(i))
      net = BatchNormLayer(net, act=lrelu,
          is_train=is_train, gamma_init=gamma_init, name='e{}/b2'.format(i))
      net = Conv2d(net, df_dim*2, (3, 3), (1, 1),
          padding='SAME', W_init=w_init, b_init=b_init, name='e{}/c3'.format(i))
      net = BatchNormLayer(net, #act=tf.nn.relu,
          is_train=is_train, gamma_init=gamma_init, name='e{}/b3'.format(i))
      net_h1 = ElementwiseLayer(layer=[net_h1, net], combine_fn=tf.add, name='e{}/add'.format(i))
      net_h1.outputs = lrelu(net_h1.outputs)

net_h2 = Conv2d(net_h1, df_dim*4, (4, 4), (2, 2), act=None,               
      padding='SAME', W_init=w_init, b_init=b_init, name='e_h2/conv2d')
net_h2 = BatchNormLayer(net_h2,                           
      is_train=is_train, gamma_init=gamma_init, name='e_h2/batchnorm')
....

[zsdonghao]

i think the problem have been solved?