diff options
| -rw-r--r-- | cli/app/search/search_class.py | 9 |
1 files changed, 5 insertions, 4 deletions
diff --git a/cli/app/search/search_class.py b/cli/app/search/search_class.py index a41a141..139ca36 100644 --- a/cli/app/search/search_class.py +++ b/cli/app/search/search_class.py @@ -100,11 +100,12 @@ def find_nearest_vector(sess, generator, opt_fp_in, opt_dims, out_images, out_la ## clip the Z encoding - opt_clip = 1.0 + opt_clip = 1.5 - clipped_encoding = tf.where(tf.abs(input_z) >= opt_clip, + clipped_latent = tf.where(tf.abs(input_z) >= opt_clip, tf.random.uniform([batch_size, z_dim], minval=-opt_clip, maxval=opt_clip), input_z) - clip_latent = tf.assign(input_z, clipped_encoding) + clipped_alpha = tf.compat.v1.placeholder(dtype=np.float32, shape=()) + clip_latent = tf.assign(input_z, clipped_latent * (1 - normalized_alpha) + input_z * normalized_alpha) ## normalize the Y encoding # normalized_labels = tf.nn.l2_normalize(input_y) @@ -189,7 +190,7 @@ def find_nearest_vector(sess, generator, opt_fp_in, opt_dims, out_images, out_la if i % 20 == 0: print('iter: {}, loss: {}'.format(i, curr_loss)) if i > 0: - if opt_stochastic_clipping and (i % opt_clip_interval) == 0: + if opt_stochastic_clipping and (i % opt_clip_interval) == 0 and i < opt_steps * 0.45: sess.run(clip_latent, { normalized_alpha: i / opt_steps }) if opt_label_clipping and (i % opt_clip_interval) == 0: sess.run(clip_labels, { normalized_alpha: i / opt_steps }) |