Internal change

official/vision/beta/projects/basnet/evaluation/mae.py

@@ -26,11 +26,7 @@ class MAE(object):
   """Mean Absolute Error(MAE) metric for basnet."""
 
   def __init__(self):
-    """Constructs MAE metric class.
-    
-    Args:
-
-    """
+    """Constructs MAE metric class."""
     self.reset_states()
 
   @property
@@ -55,7 +51,6 @@ class MAE(object):
     Returns:
       average_mae: average MAE with float numpy.
     """
-
     mae_total = 0.0
 
     for i, (true, pred) in enumerate(zip(self._groundtruths,
@@ -72,7 +67,6 @@ class MAE(object):
   def _mask_normalize(self, mask):
     return mask/(np.amax(mask)+1e-8)
 
-
   def _compute_mae(self, true, pred):
     h, w = true.shape[0], true.shape[1]
     mask1 = self._mask_normalize(true)

official/vision/beta/projects/basnet/evaluation/max_f.py

@@ -26,12 +26,7 @@ class maxFscore(object):
   """Maximum F-score metric for basnet."""
   
   def __init__(self):
-    """Constructs BASNet evaluation class.
-
-    Args:
-
-    """
-
+    """Constructs BASNet evaluation class."""
     self.reset_states()
 
   @property
@@ -76,7 +71,6 @@ class maxFscore(object):
     recalls    = np.sum(recalls,0)/(len(self._groundtruths)+1e-8)
     f          = (1+beta)*precisions*recalls/(beta*precisions+recalls+1e-8)
     f_max      = np.max(f)
-
     f_max = f_max.astype(np.float32)
 
     return f_max
@@ -128,7 +122,6 @@ class maxFscore(object):
                      groundtruth masks. range [0, 1]
       predictions  : Tuple of signle Tensor [batch, width, height, 1],
                      predicted masks. range [0, 1]
-    
     """
     groundtruths, predictions = self._convert_to_numpy(groundtruths[0],
                                                        predictions[0])

official/vision/beta/projects/basnet/evaluation/relax_f.py

@@ -12,7 +12,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # ==============================================================================
-"""Metrics for basnet"""
 
 # Import libraries
 import numpy as np
@@ -23,12 +22,7 @@ class relaxedFscore(object):
   """Relaxed F-score metric for basnet."""
 
   def __init__(self):
-    """Constructs BASNet evaluation class.
-
-    Args:
-
-    """
-
+    """Constructs BASNet evaluation class."""
     self.reset_states()
 
   @property
@@ -86,10 +80,8 @@ class relaxedFscore(object):
  
       pre, rec = self._compute_relax_pre_rec(true_xor, pred_xor, rho)
       relax_fs[i] = (1+beta)*pre*rec/(beta*pre+rec+1e-8)
-    
 
     relax_f = np.sum(relax_fs,0)/(len(self._groundtruths)+1e-8)
-    
     relax_f = relax_f.astype(np.float32)
 
     return relax_f
@@ -134,7 +126,6 @@ class relaxedFscore(object):
                      groundtruth masks. range [0, 1]
       predictions  : Tuple of single Tensor [batch, width, height, 1],
                      predicted masks. range [0, 1]
-    
     """
 
     groundtruths, predictions = self._convert_to_numpy(groundtruths[0],

official/vision/beta/projects/basnet/modeling/layers/nn_blocks.py

@@ -27,7 +27,7 @@ class ConvBlock(tf.keras.layers.Layer):
   def __init__(self,
                filters,
                strides,
-               dilation_rate,
+               dilation_rate=1,
                kernel_size=3,
                kernel_initializer='VarianceScaling',
                kernel_regularizer=None,

official/vision/beta/projects/basnet/tasks/basnet.py

@@ -46,7 +46,6 @@ def build_basnet_model(
   norm_activation_config = model_config.norm_activation
 
   decoder = basnet_decoder.BASNet_Decoder(
-        input_specs=backbone.output_specs,
         use_sync_bn=norm_activation_config.use_sync_bn,
         norm_momentum=norm_activation_config.norm_momentum,
         norm_epsilon=norm_activation_config.norm_epsilon,