[GitHub] chowkamlee81 opened a new issue #9143: maximum value of activation function using tanh goes beyoynd maximum value...

[GitBox <gi...@apache.org>]

chowkamlee81 opened a new issue #9143: maximum value of activation function using tanh goes beyoynd maximum value...
URL: https://github.com/apache/incubator-mxnet/issues/9143
 
 
   I had a network below
   
   data = mx.symbol.Variable(name="data")
       data_prevOne = mx.symbol.Variable(name="data_prevOne")
       data_prevTwo = mx.symbol.Variable(name="data_prevTwo")
       data_prevThree = mx.symbol.Variable(name="data_prevThree")
       seg_cls_gt = mx.symbol.Variable(name='label')
       data_by_2 = mx.symbol.Convolution(name='data_by_2', data=data, num_filter=1, pad=(0, 0), kernel=(1, 1),stride=(2, 2), no_bias=True)
       data_prevOne_by_2 = mx.symbol.Convolution(name='data_prevOne_by_2', data=data_prevOne, num_filter=1, pad=(0, 0),kernel=(1, 1), stride=(2, 2), no_bias=True)
       data_prevTwo_by_2 = mx.symbol.Convolution(name='data_prevTwo_by_2', data=data_prevTwo, num_filter=1, pad=(0, 0),kernel=(1, 1), stride=(2, 2), no_bias=True)
       data_prevThree_by_2 = mx.symbol.Convolution(name='data_prevThree_by_2', data=data_prevThree, num_filter=1,pad=(0, 0), kernel=(1, 1), stride=(2, 2), no_bias=True)
   
       data_by_4 = mx.symbol.Convolution(name='data_by_4', data=data_by_2, num_filter=1, pad=(0, 0), kernel=(1, 1),stride=(2, 2), no_bias=True)
       data_prevOne_by_4 = mx.symbol.Convolution(name='data_prevOne_by_4', data=data_prevOne_by_2, num_filter=1,pad=(0, 0), kernel=(1, 1), stride=(2, 2), no_bias=True)
       data_prevTwo_by_4 = mx.symbol.Convolution(name='data_prevTwo_by_4', data=data_prevTwo_by_2, num_filter=1,pad=(0, 0), kernel=(1, 1), stride=(2, 2), no_bias=True)
       data_prevThree_by_4 = mx.symbol.Convolution(name='data_prevThree_by_4', data=data_prevThree_by_2, num_filter=1,pad=(0, 0), kernel=(1, 1), stride=(2, 2), no_bias=True)
   
       data_input_scale0 = mx.symbol.Concat(data_by_4, data_prevOne_by_4, data_prevTwo_by_4,data_prevThree_by_4, dim=1)
       conv1_scale0 = mx.symbol.Convolution(name='conv1_scale0', data=data_input_scale0, num_filter=128, pad=(1, 1),kernel=(3, 3), stride=(1, 1), no_bias=True)
       conv1_scale0_relu = mx.symbol.Activation(name='conv1_scale0_relu', data=conv1_scale0, act_type='tanh')
       conv2_scale0 = mx.symbol.Convolution(name='conv2_scale0', data=conv1_scale0_relu, num_filter=256, pad=(1, 1),kernel=(3, 3), stride=(1, 1), no_bias=True)
       conv2_scale0_relu = mx.symbol.Activation(name='conv2_scale0_relu', data=conv2_scale0, act_type='tanh')
       conv3_scale0 = mx.symbol.Convolution(name='conv3_scale0', data=conv2_scale0_relu, num_filter=128, pad=(1, 1),kernel=(3, 3), stride=(1, 1), no_bias=True)
       conv3_scale0_relu = mx.symbol.Activation(name='conv3_scale0_relu', data=conv3_scale0, act_type='tanh')
       conv4_scale0 = mx.symbol.Convolution(name='conv4_scale0', data=conv3_scale0_relu, num_filter=1, pad=(1, 1),kernel=(3, 3), stride=(1, 1), no_bias=True)
       conv4_scale0_relu = mx.symbol.Activation(name='conv4_scale0_relu', data=conv4_scale0, act_type='tanh')
       conv4_scale0_relu = mx.symbol.UpSampling(conv4_scale0_relu, num_filter=1, scale=2, sample_type='bilinear',num_args=2, name="upsampling0")
   Output of  conv4_scale0_relu i has a maximum value of 2.034. How can tanh values goes beyond 1 and -1. My input dat arange is between 0 to 255.. Kindly suggest

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services