9ae3f7ec8f064f4d0fbf84d23455a8039bcbb390,shared/gan.py,,discriminator_pyramid,#Any#Any#Any#Any#Any#,395

Before Change

    filter_size_w = int(result.get_shape()[1])
    filter_size_h = int(result.get_shape()[2])
    filter = [1,filter_size_w,filter_size_h,1]
    stride = [1,filter_size_w,filter_size_h,1]
    result = tf.nn.max_pool(result, ksize=filter, strides=stride, padding="SAME")

    result = batch_norm(config["batch_size"], name="d_expand_bn_end_"+str(i))(result)
    result = activation(result)

After Change

      result = activation(result)
      // APPEND xs[i] and gs[i]
      xg = tf.concat(0, [xs[i+1], gs[i+1]])
      xg += tf.random_normal(xg.get_shape(), mean=0, stddev=config["d_noise"], dtype=config["dtype"])

      print("+++++",result, xg, "____")
      xg = conv2d(xg, 4*(i+1), name="d_add_xg"+str(i), k_w=1, k_h=1, d_h=1, d_w=1)
      xg = batch_norm(config["batch_size"], name="d_add_xg_bn_"+str(i))(xg)

Instances