up

3 files changed, 16 insertions, 10 deletions

diff --git a/cli/app/commands/biggan/extract_dense_vectors.py b/cli/app/commands/biggan/extract_dense_vectors.py
index 9f7644e..5320496 100644
--- a/cli/app/commands/biggan/extract_dense_vectors.py
+++ b/cli/app/commands/biggan/extract_dense_vectors.py
@@ -25,9 +25,11 @@ from app.search.json import params_dense_dict
   help='Feature layers used for loss')
 @click.option('-snap', '--snapshot_interval', 'opt_snapshot_interval', default=20,
   help='Interval to store sample images')
+@click.option('-clip', '--clip_interval', 'opt_clip_interval', default=500,
+  help='Interval to clip vectors')
 @click.pass_context
 def cli(ctx, opt_folder_id, opt_latent_steps, opt_dense_steps, opt_video,
-  opt_stochastic_clipping, opt_label_clipping, opt_use_feature_detector, opt_feature_layers, opt_snapshot_interval):
+  opt_stochastic_clipping, opt_label_clipping, opt_use_feature_detector, opt_feature_layers, opt_snapshot_interval, opt_clip_interval):
   """
   The full process:
   - Fetch new images from the cortex
@@ -59,7 +61,8 @@ def cli(ctx, opt_folder_id, opt_latent_steps, opt_dense_steps, opt_video,
     opt_label_clipping=opt_label_clipping,
     opt_use_feature_detector=opt_use_feature_detector,
     opt_feature_layers=opt_feature_layers,
-    opt_snapshot_interval=opt_snapshot_interval
+    opt_snapshot_interval=opt_snapshot_interval,
+    opt_clip_interval=opt_clip_interval
   )
 
   params = params_dense_dict(tag)
diff --git a/cli/app/commands/biggan/search_class.py b/cli/app/commands/biggan/search_class.py
index 6e1df95..050fbef 100644
--- a/cli/app/commands/biggan/search_class.py
+++ b/cli/app/commands/biggan/search_class.py
@@ -29,10 +29,11 @@ from app.utils.cortex_utils import cortex_folder, download_cortex_files, find_un
   help='Feature layers used for loss')
 @click.option('-snap', '--snapshot_interval', 'opt_snapshot_interval', default=20,
   help='Interval to store sample images')
-
+@click.option('-snap', '--clip_interval', 'opt_clip_interval', default=500,
+  help='Interval to clip vectors')
 @click.pass_context
 def cli(ctx, opt_folder_id, opt_fp_in, opt_dims, opt_steps, opt_limit, opt_video, opt_tag,
-  opt_stochastic_clipping, opt_label_clipping, opt_use_feature_detector, opt_feature_layers, opt_snapshot_interval):
+  opt_stochastic_clipping, opt_label_clipping, opt_use_feature_detector, opt_feature_layers, opt_snapshot_interval, opt_clip_interval):
   """
   Search for an image (class vector) in BigGAN using gradient descent
   """
@@ -59,4 +60,4 @@ def cli(ctx, opt_folder_id, opt_fp_in, opt_dims, opt_steps, opt_limit, opt_video
   opt_feature_layers = opt_feature_layers.split(',')
 
   find_nearest_vector_for_images(paths, opt_dims, opt_steps, opt_video, opt_tag, opt_limit,
-    opt_stochastic_clipping, opt_label_clipping, opt_use_feature_detector, opt_feature_layers, opt_snapshot_interval)
+    opt_stochastic_clipping, opt_label_clipping, opt_use_feature_detector, opt_feature_layers, opt_snapshot_interval, opt_clip_interval)
diff --git a/cli/app/search/search_class.py b/cli/app/search/search_class.py
index cd53a71..134a139 100644
--- a/cli/app/search/search_class.py
+++ b/cli/app/search/search_class.py
@@ -46,7 +46,7 @@ feature_layer_names = {
 
 def find_nearest_vector_for_images(paths, opt_dims, opt_steps, opt_video, opt_tag,
   opt_limit=-1, opt_stochastic_clipping=0, opt_label_clipping=0,
-  opt_use_feature_detector=False, opt_feature_layers=[1,2,4,7], opt_snapshot_interval=20):
+  opt_use_feature_detector=False, opt_feature_layers=[1,2,4,7], opt_snapshot_interval=20, opt_clip_interval=500):
   tf.reset_default_graph()
   sess = tf.compat.v1.Session()
   print("Initializing generator...")
@@ -66,13 +66,13 @@ def find_nearest_vector_for_images(paths, opt_dims, opt_steps, opt_video, opt_ta
       break
     out_fns[index] = os.path.basename(path)
     fp_frames = find_nearest_vector(sess, generator, path, opt_dims, out_images, out_labels, out_latent, opt_steps, index,
-      opt_stochastic_clipping, opt_label_clipping, opt_use_feature_detector, opt_feature_layers, opt_snapshot_interval)
+      opt_stochastic_clipping, opt_label_clipping, opt_use_feature_detector, opt_feature_layers, opt_snapshot_interval, opt_clip_interval)
     if opt_video:
       export_video(fp_frames)
   sess.close()
 
 def find_nearest_vector(sess, generator, opt_fp_in, opt_dims, out_images, out_labels, out_latent, opt_steps, index,
-  opt_stochastic_clipping, opt_label_clipping, opt_use_feature_detector, opt_feature_layers, opt_snapshot_interval):
+  opt_stochastic_clipping, opt_label_clipping, opt_use_feature_detector, opt_feature_layers, opt_snapshot_interval, opt_clip_interval):
   """
   Find the closest latent and class vectors for an image. Store the class vector in an HDF5.
   """
@@ -122,8 +122,10 @@ def find_nearest_vector(sess, generator, opt_fp_in, opt_dims, out_images, out_la
     feature_extractor = hub.Module("https://tfhub.dev/google/imagenet/inception_v3/feature_vector/1")
 
     # Convert images from range [-1, 1] channels_first to [0, 1] channels_last.
-    gen_img_1 = tf.transpose(output / 2.0 + 0.5, [0, 2, 3, 1])
-    target_img_1 = tf.transpose(target / 2.0 + 0.5, [0, 2, 3, 1])
+    # gen_img_1 = tf.transpose(output / 2.0 + 0.5, [0, 2, 3, 1])
+    # target_img_1 = tf.transpose(target / 2.0 + 0.5, [0, 2, 3, 1])
+    gen_img_1 = output / 2.0 + 0.5
+    target_img_1 = target / 2.0 + 0.5
 
     # Convert images to appropriate size for feature extraction.
     height, width = hub.get_expected_image_size(feature_extractor)