Hi, Why do not you add bias term for the conv_transpose which is in the line 51 in ops.py? `convt = tf.nn.conv2d_transpose(x, w, output_shape=outputShape, strides=[1,2,2,1])` I think it should be: `convt = tf.nn.conv2d_transpose(x, w, output_shape=outputShape, strides=[1,2,2,1]) + b`